diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..0249676495bfecb47065a34e3467c9282cb22cea
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1000,
+  "tokens_span": 18284544.0,
+  "tokens_sup": 19308544.0,
+  "examples_seen": 1024000,
+  "val_ce": 1.3160197734832764,
+  "val_ce_L20": 1.3070255517959595,
+  "val_ce_L24": 1.3439029455184937,
+  "val_ce_L28": 1.2906959056854248,
+  "val_ce_L32": 1.3591547012329102,
+  "val_ce_L36": 1.2168859243392944,
+  "val_ce_L40": 1.2491480112075806,
+  "val_ce_L44": 1.2386751174926758,
+  "val_ce_L48": 1.2798957824707031,
+  "val_ce_L52": 1.2783353328704834,
+  "val_ce_L56": 1.4324098825454712,
+  "val_ce_L60": 1.501736044883728,
+  "train_ce": 1.2460187441706658,
+  "train_val_gap": 0.07000102931261054,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..31de86a5afe40ccc5dcd006265e86968e16db535
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step1500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1500,
+  "tokens_span": 27467776.0,
+  "tokens_sup": 29003776.0,
+  "examples_seen": 1536000,
+  "val_ce": 1.2897164821624756,
+  "val_ce_L20": 1.2548513412475586,
+  "val_ce_L24": 1.3059653043746948,
+  "val_ce_L28": 1.2445114850997925,
+  "val_ce_L32": 1.3372445106506348,
+  "val_ce_L36": 1.1796061992645264,
+  "val_ce_L40": 1.2273386716842651,
+  "val_ce_L44": 1.210042119026184,
+  "val_ce_L48": 1.2801361083984375,
+  "val_ce_L52": 1.2656112909317017,
+  "val_ce_L56": 1.4198745489120483,
+  "val_ce_L60": 1.488851547241211,
+  "train_ce": 1.1425579063892364,
+  "train_val_gap": 0.1471585757732392,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..3c39381fefe83feb2940a3e6126068af245abc3b
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2000,
+  "tokens_span": 36853760.0,
+  "tokens_sup": 38901760.0,
+  "examples_seen": 2048000,
+  "val_ce": 1.2604827880859375,
+  "val_ce_L20": 1.210369348526001,
+  "val_ce_L24": 1.2374399900436401,
+  "val_ce_L28": 1.1942862272262573,
+  "val_ce_L32": 1.3073383569717407,
+  "val_ce_L36": 1.1571012735366821,
+  "val_ce_L40": 1.2083390951156616,
+  "val_ce_L44": 1.188383936882019,
+  "val_ce_L48": 1.2531511783599854,
+  "val_ce_L52": 1.248670220375061,
+  "val_ce_L56": 1.4093008041381836,
+  "val_ce_L60": 1.4756282567977905,
+  "train_ce": 1.0634016906917094,
+  "train_val_gap": 0.19708109739422808,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..44a5f5ac0f39775014cd89cc4a20851a314b8b15
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step2500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2500,
+  "tokens_span": 45751296.0,
+  "tokens_sup": 48311296.0,
+  "examples_seen": 2560000,
+  "val_ce": 1.287048101425171,
+  "val_ce_L20": 1.2432713508605957,
+  "val_ce_L24": 1.2799700498580933,
+  "val_ce_L28": 1.196609616279602,
+  "val_ce_L32": 1.3380208015441895,
+  "val_ce_L36": 1.1756632328033447,
+  "val_ce_L40": 1.259678840637207,
+  "val_ce_L44": 1.2235233783721924,
+  "val_ce_L48": 1.2836198806762695,
+  "val_ce_L52": 1.2515300512313843,
+  "val_ce_L56": 1.4206113815307617,
+  "val_ce_L60": 1.507584571838379,
+  "train_ce": 0.9606544596254826,
+  "train_val_gap": 0.3263936417996883,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..442ad7d25d19b66a1572e5696df07b03c6de3d06
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3000,
+  "tokens_span": 55110656.0,
+  "tokens_sup": 58182656.0,
+  "examples_seen": 3072000,
+  "val_ce": 1.2937355041503906,
+  "val_ce_L20": 1.2337371110916138,
+  "val_ce_L24": 1.2826884984970093,
+  "val_ce_L28": 1.2169009447097778,
+  "val_ce_L32": 1.3512821197509766,
+  "val_ce_L36": 1.2057968378067017,
+  "val_ce_L40": 1.2689497470855713,
+  "val_ce_L44": 1.1988202333450317,
+  "val_ce_L48": 1.3063496351242065,
+  "val_ce_L52": 1.268314003944397,
+  "val_ce_L56": 1.413094162940979,
+  "val_ce_L60": 1.5060462951660156,
+  "train_ce": 0.9122910050153732,
+  "train_val_gap": 0.38144449913501743,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..ccc7b81add5eda241d1e52506c2b8a4b532ccaa5
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step3221/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3221,
+  "tokens_span": 58909696.0,
+  "tokens_sup": 62208000.0,
+  "examples_seen": 3298304,
+  "val_ce": 1.3266960382461548,
+  "val_ce_L20": 1.250138282775879,
+  "val_ce_L24": 1.3243069648742676,
+  "val_ce_L28": 1.2651358842849731,
+  "val_ce_L32": 1.379777431488037,
+  "val_ce_L36": 1.210906744003296,
+  "val_ce_L40": 1.2793513536453247,
+  "val_ce_L44": 1.2760872840881348,
+  "val_ce_L48": 1.3282365798950195,
+  "val_ce_L52": 1.2896722555160522,
+  "val_ce_L56": 1.4525333642959595,
+  "val_ce_L60": 1.564955711364746,
+  "train_ce": 0.8271289839329223,
+  "train_val_gap": 0.49956705431323245,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..f585701996b4f48d10c4bc1351a472c348d4dc36
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend/step500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 500,
+  "tokens_span": 9432064.0,
+  "tokens_sup": 9944064.0,
+  "examples_seen": 512000,
+  "val_ce": 1.4270954132080078,
+  "val_ce_L20": 1.4415392875671387,
+  "val_ce_L24": 1.4663161039352417,
+  "val_ce_L28": 1.4298101663589478,
+  "val_ce_L32": 1.4932904243469238,
+  "val_ce_L36": 1.347434639930725,
+  "val_ce_L40": 1.340230107307434,
+  "val_ce_L44": 1.3578640222549438,
+  "val_ce_L48": 1.36846923828125,
+  "val_ce_L52": 1.3750728368759155,
+  "val_ce_L56": 1.5103765726089478,
+  "val_ce_L60": 1.585629940032959,
+  "train_ce": 1.6530334638357163,
+  "train_val_gap": -0.22593805062770844,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..46298e6b8a86055f52ce77f5eede437e9ce92f08
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1000,
+  "tokens_span": 18199552.0,
+  "tokens_sup": 19223552.0,
+  "examples_seen": 1024000,
+  "val_ce": 1.2794396877288818,
+  "val_ce_L20": 1.2609106302261353,
+  "val_ce_L24": 1.2861063480377197,
+  "val_ce_L28": 1.2240310907363892,
+  "val_ce_L32": 1.347455620765686,
+  "val_ce_L36": 1.1587861776351929,
+  "val_ce_L40": 1.2171308994293213,
+  "val_ce_L44": 1.2146910429000854,
+  "val_ce_L48": 1.2505067586898804,
+  "val_ce_L52": 1.26174795627594,
+  "val_ce_L56": 1.4096708297729492,
+  "val_ce_L60": 1.4684720039367676,
+  "train_ce": 1.2844150429964065,
+  "train_val_gap": -0.004975355267524684,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..35eca2b037dac1d82f01a35bf878a5fc629a4501
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 10000,
+  "tokens_span": 182770688.0,
+  "tokens_sup": 193010688.0,
+  "examples_seen": 10240000,
+  "val_ce": 1.147078514099121,
+  "val_ce_L20": 1.057914137840271,
+  "val_ce_L24": 1.1253106594085693,
+  "val_ce_L28": 1.0774312019348145,
+  "val_ce_L32": 1.1754080057144165,
+  "val_ce_L36": 1.0492713451385498,
+  "val_ce_L40": 1.120781660079956,
+  "val_ce_L44": 1.0773696899414062,
+  "val_ce_L48": 1.172749638557434,
+  "val_ce_L52": 1.1298480033874512,
+  "val_ce_L56": 1.2743521928787231,
+  "val_ce_L60": 1.382391333580017,
+  "train_ce": 0.8972788824960589,
+  "train_val_gap": 0.24979963160306218,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..33029d961cb0b7db49337b57ddfd16194d37ad51
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 10500,
+  "tokens_span": 191866880.0,
+  "tokens_sup": 202618880.0,
+  "examples_seen": 10752000,
+  "val_ce": 1.1499695777893066,
+  "val_ce_L20": 1.0618640184402466,
+  "val_ce_L24": 1.1147408485412598,
+  "val_ce_L28": 1.0773769617080688,
+  "val_ce_L32": 1.180230736732483,
+  "val_ce_L36": 1.0658936500549316,
+  "val_ce_L40": 1.1348178386688232,
+  "val_ce_L44": 1.0911335945129395,
+  "val_ce_L48": 1.1644901037216187,
+  "val_ce_L52": 1.1320792436599731,
+  "val_ce_L56": 1.2735570669174194,
+  "val_ce_L60": 1.3724385499954224,
+  "train_ce": 0.8760272574573755,
+  "train_val_gap": 0.2739423203319311,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..541361d21bea644ec0abfa7c9942fa92f3b7f07f
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step10603/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 10603,
+  "tokens_span": 193746944.0,
+  "tokens_sup": 204604416.0,
+  "examples_seen": 10857472,
+  "val_ce": 1.1377981901168823,
+  "val_ce_L20": 1.0405043363571167,
+  "val_ce_L24": 1.1026939153671265,
+  "val_ce_L28": 1.0527781248092651,
+  "val_ce_L32": 1.185759425163269,
+  "val_ce_L36": 1.0371209383010864,
+  "val_ce_L40": 1.1009219884872437,
+  "val_ce_L44": 1.0850839614868164,
+  "val_ce_L48": 1.1659104824066162,
+  "val_ce_L52": 1.1336084604263306,
+  "val_ce_L56": 1.2689943313598633,
+  "val_ce_L60": 1.3700145483016968,
+  "train_ce": 0.8688119102305579,
+  "train_val_gap": 0.2689862798863244,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..7e7a02595ca9fb51165da2521d9a8e46e138d965
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step1500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1500,
+  "tokens_span": 27310080.0,
+  "tokens_sup": 28846080.0,
+  "examples_seen": 1536000,
+  "val_ce": 1.2267848253250122,
+  "val_ce_L20": 1.1825789213180542,
+  "val_ce_L24": 1.208807349205017,
+  "val_ce_L28": 1.1780914068222046,
+  "val_ce_L32": 1.2686996459960938,
+  "val_ce_L36": 1.1071984767913818,
+  "val_ce_L40": 1.1559374332427979,
+  "val_ce_L44": 1.1627734899520874,
+  "val_ce_L48": 1.214462161064148,
+  "val_ce_L52": 1.224327802658081,
+  "val_ce_L56": 1.3841696977615356,
+  "val_ce_L60": 1.4372719526290894,
+  "train_ce": 1.2000047302246093,
+  "train_val_gap": 0.026780095100402912,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..7752839bcff8800bfa4ca9f0953d39a017ed6882
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2000,
+  "tokens_span": 36663296.0,
+  "tokens_sup": 38711296.0,
+  "examples_seen": 2048000,
+  "val_ce": 1.2051620483398438,
+  "val_ce_L20": 1.1682356595993042,
+  "val_ce_L24": 1.1859829425811768,
+  "val_ce_L28": 1.1621588468551636,
+  "val_ce_L32": 1.2459697723388672,
+  "val_ce_L36": 1.088072419166565,
+  "val_ce_L40": 1.1438509225845337,
+  "val_ce_L44": 1.1326831579208374,
+  "val_ce_L48": 1.2070726156234741,
+  "val_ce_L52": 1.1805998086929321,
+  "val_ce_L56": 1.3471125364303589,
+  "val_ce_L60": 1.4250956773757935,
+  "train_ce": 1.1669836622178555,
+  "train_val_gap": 0.038178386121988206,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..ba3aa19b4f437c6ce7dea61fc8e67e18156d7cc0
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step2500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2500,
+  "tokens_span": 45873152.0,
+  "tokens_sup": 48433152.0,
+  "examples_seen": 2560000,
+  "val_ce": 1.2405095100402832,
+  "val_ce_L20": 1.214521050453186,
+  "val_ce_L24": 1.2365515232086182,
+  "val_ce_L28": 1.1566509008407593,
+  "val_ce_L32": 1.2783430814743042,
+  "val_ce_L36": 1.1128908395767212,
+  "val_ce_L40": 1.1881002187728882,
+  "val_ce_L44": 1.1614257097244263,
+  "val_ce_L48": 1.222954511642456,
+  "val_ce_L52": 1.2283871173858643,
+  "val_ce_L56": 1.3860965967178345,
+  "val_ce_L60": 1.4867610931396484,
+  "train_ce": 1.1127116147279739,
+  "train_val_gap": 0.12779789531230934,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..e03d4b2e424ac31bb20a2face1c1cb64529f388e
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3000,
+  "tokens_span": 55260160.0,
+  "tokens_sup": 58332160.0,
+  "examples_seen": 3072000,
+  "val_ce": 1.1841005086898804,
+  "val_ce_L20": 1.1110998392105103,
+  "val_ce_L24": 1.1693896055221558,
+  "val_ce_L28": 1.1327086687088013,
+  "val_ce_L32": 1.2244523763656616,
+  "val_ce_L36": 1.0591293573379517,
+  "val_ce_L40": 1.1514557600021362,
+  "val_ce_L44": 1.0979892015457153,
+  "val_ce_L48": 1.2017370462417603,
+  "val_ce_L52": 1.1796592473983765,
+  "val_ce_L56": 1.326141595840454,
+  "val_ce_L60": 1.4005502462387085,
+  "train_ce": 1.088088481605053,
+  "train_val_gap": 0.09601202708482748,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..70eaaf8b5ca76e743a9132a7ea7f7f6299b22bdf
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step3500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3500,
+  "tokens_span": 64128000.0,
+  "tokens_sup": 67712000.0,
+  "examples_seen": 3584000,
+  "val_ce": 1.1771403551101685,
+  "val_ce_L20": 1.1341434717178345,
+  "val_ce_L24": 1.1635295152664185,
+  "val_ce_L28": 1.1177664995193481,
+  "val_ce_L32": 1.203652024269104,
+  "val_ce_L36": 1.0604240894317627,
+  "val_ce_L40": 1.1395676136016846,
+  "val_ce_L44": 1.1139315366744995,
+  "val_ce_L48": 1.1645500659942627,
+  "val_ce_L52": 1.160385251045227,
+  "val_ce_L56": 1.3070547580718994,
+  "val_ce_L60": 1.4071545600891113,
+  "train_ce": 1.0349715248048306,
+  "train_val_gap": 0.1421688303053379,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..318c99ad8b7d21132fea5ca57555fcb651c33066
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 4000,
+  "tokens_span": 73389056.0,
+  "tokens_sup": 77485056.0,
+  "examples_seen": 4096000,
+  "val_ce": 1.1638338565826416,
+  "val_ce_L20": 1.107223391532898,
+  "val_ce_L24": 1.1191350221633911,
+  "val_ce_L28": 1.1062175035476685,
+  "val_ce_L32": 1.1980808973312378,
+  "val_ce_L36": 1.0456444025039673,
+  "val_ce_L40": 1.1256791353225708,
+  "val_ce_L44": 1.082163691520691,
+  "val_ce_L48": 1.1732081174850464,
+  "val_ce_L52": 1.1534826755523682,
+  "val_ce_L56": 1.302382469177246,
+  "val_ce_L60": 1.4165762662887573,
+  "train_ce": 1.0339341437369585,
+  "train_val_gap": 0.12989971284568314,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..2bf9ca819eb1e65161bcd1987cc13dc989b2f95d
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step4500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 4500,
+  "tokens_span": 82491392.0,
+  "tokens_sup": 87099392.0,
+  "examples_seen": 4608000,
+  "val_ce": 1.1488654613494873,
+  "val_ce_L20": 1.0961030721664429,
+  "val_ce_L24": 1.1312881708145142,
+  "val_ce_L28": 1.0962066650390625,
+  "val_ce_L32": 1.1835967302322388,
+  "val_ce_L36": 1.0442619323730469,
+  "val_ce_L40": 1.100257158279419,
+  "val_ce_L44": 1.0577815771102905,
+  "val_ce_L48": 1.1531201601028442,
+  "val_ce_L52": 1.1424529552459717,
+  "val_ce_L56": 1.2826164960861206,
+  "val_ce_L60": 1.3760792016983032,
+  "train_ce": 1.0117913684397937,
+  "train_val_gap": 0.13707409290969363,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..a34ca9389cf52271323ad2134717b2a9170b6a05
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 500,
+  "tokens_span": 8822784.0,
+  "tokens_sup": 9334784.0,
+  "examples_seen": 512000,
+  "val_ce": 1.366627812385559,
+  "val_ce_L20": 1.3562078475952148,
+  "val_ce_L24": 1.4027763605117798,
+  "val_ce_L28": 1.3617277145385742,
+  "val_ce_L32": 1.428831696510315,
+  "val_ce_L36": 1.246384620666504,
+  "val_ce_L40": 1.2792402505874634,
+  "val_ce_L44": 1.2863187789916992,
+  "val_ce_L48": 1.3192262649536133,
+  "val_ce_L52": 1.3442038297653198,
+  "val_ce_L56": 1.4806443452835083,
+  "val_ce_L60": 1.5540486574172974,
+  "train_ce": 1.6324574474096298,
+  "train_val_gap": -0.26582963502407075,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..efa33183f0fac0463a204f3d8e67cf39fbdb67cf
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 5000,
+  "tokens_span": 91223040.0,
+  "tokens_sup": 96343040.0,
+  "examples_seen": 5120000,
+  "val_ce": 1.1533342599868774,
+  "val_ce_L20": 1.0859169960021973,
+  "val_ce_L24": 1.1450446844100952,
+  "val_ce_L28": 1.0901325941085815,
+  "val_ce_L32": 1.1886564493179321,
+  "val_ce_L36": 1.0375934839248657,
+  "val_ce_L40": 1.1160914897918701,
+  "val_ce_L44": 1.079738736152649,
+  "val_ce_L48": 1.1487176418304443,
+  "val_ce_L52": 1.1390045881271362,
+  "val_ce_L56": 1.2917474508285522,
+  "val_ce_L60": 1.388761043548584,
+  "train_ce": 0.9694670126587153,
+  "train_val_gap": 0.18386724732816218,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..446575875d690a6d6f84e143e06f8f97f82c6a86
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step5500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 5500,
+  "tokens_span": 100399104.0,
+  "tokens_sup": 106031104.0,
+  "examples_seen": 5632000,
+  "val_ce": 1.152165412902832,
+  "val_ce_L20": 1.095288634300232,
+  "val_ce_L24": 1.1384658813476562,
+  "val_ce_L28": 1.084311842918396,
+  "val_ce_L32": 1.1839860677719116,
+  "val_ce_L36": 1.0555185079574585,
+  "val_ce_L40": 1.1191390752792358,
+  "val_ce_L44": 1.059996485710144,
+  "val_ce_L48": 1.1529600620269775,
+  "val_ce_L52": 1.1218820810317993,
+  "val_ce_L56": 1.2884992361068726,
+  "val_ce_L60": 1.3955168724060059,
+  "train_ce": 0.9722878126204014,
+  "train_val_gap": 0.1798776002824306,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..39cc0f3aad9dabb78bbb85de7cd0328e42783aba
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 6000,
+  "tokens_span": 109400064.0,
+  "tokens_sup": 115544064.0,
+  "examples_seen": 6144000,
+  "val_ce": 1.1375170946121216,
+  "val_ce_L20": 1.0558189153671265,
+  "val_ce_L24": 1.1205387115478516,
+  "val_ce_L28": 1.0682767629623413,
+  "val_ce_L32": 1.1599210500717163,
+  "val_ce_L36": 1.0356504917144775,
+  "val_ce_L40": 1.104779601097107,
+  "val_ce_L44": 1.0510154962539673,
+  "val_ce_L48": 1.1581979990005493,
+  "val_ce_L52": 1.1190425157546997,
+  "val_ce_L56": 1.2934595346450806,
+  "val_ce_L60": 1.371867060661316,
+  "train_ce": 0.9578785347044468,
+  "train_val_gap": 0.17963855990767474,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..4bed061c0be3de5146c5505f5c701e6d64d921ba
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step6500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 6500,
+  "tokens_span": 118416384.0,
+  "tokens_sup": 125072384.0,
+  "examples_seen": 6656000,
+  "val_ce": 1.1464600563049316,
+  "val_ce_L20": 1.03939688205719,
+  "val_ce_L24": 1.1319929361343384,
+  "val_ce_L28": 1.0960053205490112,
+  "val_ce_L32": 1.1767438650131226,
+  "val_ce_L36": 1.047053337097168,
+  "val_ce_L40": 1.1164554357528687,
+  "val_ce_L44": 1.0663000345230103,
+  "val_ce_L48": 1.162348985671997,
+  "val_ce_L52": 1.131583333015442,
+  "val_ce_L56": 1.2862151861190796,
+  "val_ce_L60": 1.381182074546814,
+  "train_ce": 0.9415584835261106,
+  "train_val_gap": 0.20490157277882104,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..6d7510c8a0ea2c014d96587b0b91659017886979
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 7000,
+  "tokens_span": 127619072.0,
+  "tokens_sup": 134787072.0,
+  "examples_seen": 7168000,
+  "val_ce": 1.152179479598999,
+  "val_ce_L20": 1.0604379177093506,
+  "val_ce_L24": 1.106768250465393,
+  "val_ce_L28": 1.0734964609146118,
+  "val_ce_L32": 1.1911886930465698,
+  "val_ce_L36": 1.0541056394577026,
+  "val_ce_L40": 1.1170437335968018,
+  "val_ce_L44": 1.0637906789779663,
+  "val_ce_L48": 1.1750733852386475,
+  "val_ce_L52": 1.1379050016403198,
+  "val_ce_L56": 1.313855528831482,
+  "val_ce_L60": 1.406324863433838,
+  "train_ce": 0.9409561244100332,
+  "train_val_gap": 0.21122335518896584,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c01a299cd3131b6b05ac6cc79a33b0629b348e9e
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step7500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 7500,
+  "tokens_span": 136687616.0,
+  "tokens_sup": 144367616.0,
+  "examples_seen": 7680000,
+  "val_ce": 1.14952552318573,
+  "val_ce_L20": 1.0647298097610474,
+  "val_ce_L24": 1.1207385063171387,
+  "val_ce_L28": 1.0759092569351196,
+  "val_ce_L32": 1.1950126886367798,
+  "val_ce_L36": 1.053719401359558,
+  "val_ce_L40": 1.1120632886886597,
+  "val_ce_L44": 1.0537322759628296,
+  "val_ce_L48": 1.1645758152008057,
+  "val_ce_L52": 1.135341763496399,
+  "val_ce_L56": 1.2970000505447388,
+  "val_ce_L60": 1.3966779708862305,
+  "train_ce": 0.9238434360176325,
+  "train_val_gap": 0.22568208716809746,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..236ff9bdcbf3c45fad8a583b666db3e96982d915
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 8000,
+  "tokens_span": 145676288.0,
+  "tokens_sup": 153868288.0,
+  "examples_seen": 8192000,
+  "val_ce": 1.1315057277679443,
+  "val_ce_L20": 1.0395654439926147,
+  "val_ce_L24": 1.1090551614761353,
+  "val_ce_L28": 1.0650746822357178,
+  "val_ce_L32": 1.162055253982544,
+  "val_ce_L36": 1.028814435005188,
+  "val_ce_L40": 1.1022850275039673,
+  "val_ce_L44": 1.0509812831878662,
+  "val_ce_L48": 1.1686099767684937,
+  "val_ce_L52": 1.115482211112976,
+  "val_ce_L56": 1.274613857269287,
+  "val_ce_L60": 1.3581939935684204,
+  "train_ce": 0.9131627733260393,
+  "train_val_gap": 0.218342954441905,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..7b8de52aeebdab6af7e3f71db32bc5c2ad93a591
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step8500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 8500,
+  "tokens_span": 154830848.0,
+  "tokens_sup": 163534848.0,
+  "examples_seen": 8704000,
+  "val_ce": 1.128859519958496,
+  "val_ce_L20": 1.0483561754226685,
+  "val_ce_L24": 1.0930554866790771,
+  "val_ce_L28": 1.0353907346725464,
+  "val_ce_L32": 1.1555641889572144,
+  "val_ce_L36": 1.0134161710739136,
+  "val_ce_L40": 1.0969985723495483,
+  "val_ce_L44": 1.0533570051193237,
+  "val_ce_L48": 1.1801855564117432,
+  "val_ce_L52": 1.1179574728012085,
+  "val_ce_L56": 1.2809005975723267,
+  "val_ce_L60": 1.3753787279129028,
+  "train_ce": 0.900143549606204,
+  "train_val_gap": 0.22871597035229208,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..8a43df87d23c61902bfaacdd67d6e410c878e9c3
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 9000,
+  "tokens_span": 164231168.0,
+  "tokens_sup": 173447168.0,
+  "examples_seen": 9216000,
+  "val_ce": 1.1365206241607666,
+  "val_ce_L20": 1.0455658435821533,
+  "val_ce_L24": 1.095973253250122,
+  "val_ce_L28": 1.0792893171310425,
+  "val_ce_L32": 1.1667861938476562,
+  "val_ce_L36": 1.045080542564392,
+  "val_ce_L40": 1.107672929763794,
+  "val_ce_L44": 1.0664793252944946,
+  "val_ce_L48": 1.16043221950531,
+  "val_ce_L52": 1.108710765838623,
+  "val_ce_L56": 1.2693443298339844,
+  "val_ce_L60": 1.3804250955581665,
+  "train_ce": 0.9035262315273285,
+  "train_val_gap": 0.23299439263343813,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..07fc91a6114606e17e51b906beb1a05528088b2b
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.e1cend_full/step9500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 9500,
+  "tokens_span": 173485056.0,
+  "tokens_sup": 183213056.0,
+  "examples_seen": 9728000,
+  "val_ce": 1.1390546560287476,
+  "val_ce_L20": 1.0724434852600098,
+  "val_ce_L24": 1.099460244178772,
+  "val_ce_L28": 1.0627129077911377,
+  "val_ce_L32": 1.1767287254333496,
+  "val_ce_L36": 1.0455232858657837,
+  "val_ce_L40": 1.089565396308899,
+  "val_ce_L44": 1.0590927600860596,
+  "val_ce_L48": 1.1748243570327759,
+  "val_ce_L52": 1.1385489702224731,
+  "val_ce_L56": 1.2734483480453491,
+  "val_ce_L60": 1.3670507669448853,
+  "train_ce": 0.8905328354910016,
+  "train_val_gap": 0.24852182053774596,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 115.07945174569515,
+      "24": 115.07945174569515,
+      "28": 115.07945174569515,
+      "32": 115.07945174569515,
+      "36": 115.07945174569515,
+      "40": 115.07945174569515,
+      "44": 115.07945174569515,
+      "48": 115.07945174569515,
+      "52": 115.07945174569515,
+      "56": 115.07945174569515,
+      "60": 115.07945174569515
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.e1c/end",
+      "target_prefix": "",
+      "global_scale": "115.07945174569515",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.mean.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..769d2a2470c54fc96b26dc724b66f3b10bfb3d0c
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1000,
+  "tokens_span": 18284544.0,
+  "tokens_sup": 19308544.0,
+  "examples_seen": 1024000,
+  "val_ce": 1.3067822456359863,
+  "val_ce_L20": 1.311010718345642,
+  "val_ce_L24": 1.3134099245071411,
+  "val_ce_L28": 1.3291466236114502,
+  "val_ce_L32": 1.3413268327713013,
+  "val_ce_L36": 1.2041314840316772,
+  "val_ce_L40": 1.2423099279403687,
+  "val_ce_L44": 1.2132036685943604,
+  "val_ce_L48": 1.269819974899292,
+  "val_ce_L52": 1.2571970224380493,
+  "val_ce_L56": 1.432949185371399,
+  "val_ce_L60": 1.481791615486145,
+  "train_ce": 1.240101599395275,
+  "train_val_gap": 0.06668064624071124,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 125.65015648212251,
+      "24": 125.65015648212251,
+      "28": 125.65015648212251,
+      "32": 125.65015648212251,
+      "36": 125.65015648212251,
+      "40": 125.65015648212251,
+      "44": 125.65015648212251,
+      "48": 125.65015648212251,
+      "52": 125.65015648212251,
+      "56": 125.65015648212251,
+      "60": 125.65015648212251
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/end",
+      "target_prefix": "",
+      "global_scale": "125.65015648212251",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..183802db15244f356a7aafd608aeeaa0d5a861a1
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step1500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1500,
+  "tokens_span": 27467776.0,
+  "tokens_sup": 29003776.0,
+  "examples_seen": 1536000,
+  "val_ce": 1.281907558441162,
+  "val_ce_L20": 1.2482050657272339,
+  "val_ce_L24": 1.2905287742614746,
+  "val_ce_L28": 1.3004095554351807,
+  "val_ce_L32": 1.3180091381072998,
+  "val_ce_L36": 1.16341233253479,
+  "val_ce_L40": 1.210866093635559,
+  "val_ce_L44": 1.2067228555679321,
+  "val_ce_L48": 1.2640489339828491,
+  "val_ce_L52": 1.2305320501327515,
+  "val_ce_L56": 1.4094427824020386,
+  "val_ce_L60": 1.487418532371521,
+  "train_ce": 1.1337248183488846,
+  "train_val_gap": 0.1481827400922775,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 125.65015648212251,
+      "24": 125.65015648212251,
+      "28": 125.65015648212251,
+      "32": 125.65015648212251,
+      "36": 125.65015648212251,
+      "40": 125.65015648212251,
+      "44": 125.65015648212251,
+      "48": 125.65015648212251,
+      "52": 125.65015648212251,
+      "56": 125.65015648212251,
+      "60": 125.65015648212251
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/end",
+      "target_prefix": "",
+      "global_scale": "125.65015648212251",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..7465802aa826e099793761fe88bcec4d5c0b80cf
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2000,
+  "tokens_span": 36853760.0,
+  "tokens_sup": 38901760.0,
+  "examples_seen": 2048000,
+  "val_ce": 1.2516825199127197,
+  "val_ce_L20": 1.2152212858200073,
+  "val_ce_L24": 1.2486907243728638,
+  "val_ce_L28": 1.2567554712295532,
+  "val_ce_L32": 1.299862265586853,
+  "val_ce_L36": 1.127233624458313,
+  "val_ce_L40": 1.1916744709014893,
+  "val_ce_L44": 1.1757298707962036,
+  "val_ce_L48": 1.2070177793502808,
+  "val_ce_L52": 1.205647587776184,
+  "val_ce_L56": 1.3957041501998901,
+  "val_ce_L60": 1.4682703018188477,
+  "train_ce": 1.0533841777741908,
+  "train_val_gap": 0.19829834213852893,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 125.65015648212251,
+      "24": 125.65015648212251,
+      "28": 125.65015648212251,
+      "32": 125.65015648212251,
+      "36": 125.65015648212251,
+      "40": 125.65015648212251,
+      "44": 125.65015648212251,
+      "48": 125.65015648212251,
+      "52": 125.65015648212251,
+      "56": 125.65015648212251,
+      "60": 125.65015648212251
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/end",
+      "target_prefix": "",
+      "global_scale": "125.65015648212251",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d68b502f27b9eca3c3aaf2976f0172388557401
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step2500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2500,
+  "tokens_span": 45751296.0,
+  "tokens_sup": 48311296.0,
+  "examples_seen": 2560000,
+  "val_ce": 1.263725996017456,
+  "val_ce_L20": 1.2179388999938965,
+  "val_ce_L24": 1.2701730728149414,
+  "val_ce_L28": 1.2416613101959229,
+  "val_ce_L32": 1.3391462564468384,
+  "val_ce_L36": 1.1324636936187744,
+  "val_ce_L40": 1.2056541442871094,
+  "val_ce_L44": 1.1677128076553345,
+  "val_ce_L48": 1.243288516998291,
+  "val_ce_L52": 1.216583490371704,
+  "val_ce_L56": 1.4151073694229126,
+  "val_ce_L60": 1.4794038534164429,
+  "train_ce": 0.950124235779047,
+  "train_val_gap": 0.313601760238409,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 125.65015648212251,
+      "24": 125.65015648212251,
+      "28": 125.65015648212251,
+      "32": 125.65015648212251,
+      "36": 125.65015648212251,
+      "40": 125.65015648212251,
+      "44": 125.65015648212251,
+      "48": 125.65015648212251,
+      "52": 125.65015648212251,
+      "56": 125.65015648212251,
+      "60": 125.65015648212251
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/end",
+      "target_prefix": "",
+      "global_scale": "125.65015648212251",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..27097e8454fd2679d2a6f39388c1abb2e7649f7e
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3000,
+  "tokens_span": 55110656.0,
+  "tokens_sup": 58182656.0,
+  "examples_seen": 3072000,
+  "val_ce": 1.2769320011138916,
+  "val_ce_L20": 1.2319997549057007,
+  "val_ce_L24": 1.29216468334198,
+  "val_ce_L28": 1.2599979639053345,
+  "val_ce_L32": 1.3205012083053589,
+  "val_ce_L36": 1.1675447225570679,
+  "val_ce_L40": 1.2240315675735474,
+  "val_ce_L44": 1.1854323148727417,
+  "val_ce_L48": 1.2510474920272827,
+  "val_ce_L52": 1.2308748960494995,
+  "val_ce_L56": 1.4171429872512817,
+  "val_ce_L60": 1.4882488250732422,
+  "train_ce": 0.9036104689836502,
+  "train_val_gap": 0.37332153213024144,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 125.65015648212251,
+      "24": 125.65015648212251,
+      "28": 125.65015648212251,
+      "32": 125.65015648212251,
+      "36": 125.65015648212251,
+      "40": 125.65015648212251,
+      "44": 125.65015648212251,
+      "48": 125.65015648212251,
+      "52": 125.65015648212251,
+      "56": 125.65015648212251,
+      "60": 125.65015648212251
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/end",
+      "target_prefix": "",
+      "global_scale": "125.65015648212251",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..35dfe3eb71e21427f9c9184e0b15fa3b7f19f6f2
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step3221/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3221,
+  "tokens_span": 58909696.0,
+  "tokens_sup": 62208000.0,
+  "examples_seen": 3298304,
+  "val_ce": 1.3034698963165283,
+  "val_ce_L20": 1.2402743101119995,
+  "val_ce_L24": 1.2989715337753296,
+  "val_ce_L28": 1.3078402280807495,
+  "val_ce_L32": 1.354288101196289,
+  "val_ce_L36": 1.1720935106277466,
+  "val_ce_L40": 1.2253717184066772,
+  "val_ce_L44": 1.234476089477539,
+  "val_ce_L48": 1.2974069118499756,
+  "val_ce_L52": 1.246604323387146,
+  "val_ce_L56": 1.4470409154891968,
+  "val_ce_L60": 1.5477029085159302,
+  "train_ce": 0.8183276480341929,
+  "train_val_gap": 0.48514224828233543,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 125.65015648212251,
+      "24": 125.65015648212251,
+      "28": 125.65015648212251,
+      "32": 125.65015648212251,
+      "36": 125.65015648212251,
+      "40": 125.65015648212251,
+      "44": 125.65015648212251,
+      "48": 125.65015648212251,
+      "52": 125.65015648212251,
+      "56": 125.65015648212251,
+      "60": 125.65015648212251
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/end",
+      "target_prefix": "",
+      "global_scale": "125.65015648212251",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..9e55528f0c3b219c126b82caa63fa8b304641f75
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cend/step500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 500,
+  "tokens_span": 9432064.0,
+  "tokens_sup": 9944064.0,
+  "examples_seen": 512000,
+  "val_ce": 1.410858392715454,
+  "val_ce_L20": 1.4278206825256348,
+  "val_ce_L24": 1.4401898384094238,
+  "val_ce_L28": 1.4613767862319946,
+  "val_ce_L32": 1.4719866514205933,
+  "val_ce_L36": 1.3218529224395752,
+  "val_ce_L40": 1.3183752298355103,
+  "val_ce_L44": 1.3175454139709473,
+  "val_ce_L48": 1.3545570373535156,
+  "val_ce_L52": 1.3480076789855957,
+  "val_ce_L56": 1.5098838806152344,
+  "val_ce_L60": 1.568711280822754,
+  "train_ce": 1.6471771264076234,
+  "train_val_gap": -0.23631873369216927,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 125.65015648212251,
+      "24": 125.65015648212251,
+      "28": 125.65015648212251,
+      "32": 125.65015648212251,
+      "36": 125.65015648212251,
+      "40": 125.65015648212251,
+      "44": 125.65015648212251,
+      "48": 125.65015648212251,
+      "52": 125.65015648212251,
+      "56": 125.65015648212251,
+      "60": 125.65015648212251
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/end",
+      "target_prefix": "",
+      "global_scale": "125.65015648212251",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..7bc08f62fb58152a55b1dc9f5e431b5213b25fd7
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1000,
+  "tokens_span": 18284544.0,
+  "tokens_sup": 19308544.0,
+  "examples_seen": 1024000,
+  "val_ce": 1.3033943176269531,
+  "val_ce_L20": 1.278738021850586,
+  "val_ce_L24": 1.3009424209594727,
+  "val_ce_L28": 1.30355966091156,
+  "val_ce_L32": 1.31302011013031,
+  "val_ce_L36": 1.1875684261322021,
+  "val_ce_L40": 1.2412363290786743,
+  "val_ce_L44": 1.227853775024414,
+  "val_ce_L48": 1.28496515750885,
+  "val_ce_L52": 1.2656961679458618,
+  "val_ce_L56": 1.4421323537826538,
+  "val_ce_L60": 1.518215298652649,
+  "train_ce": 1.238468832731247,
+  "train_val_gap": 0.06492548489570615,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cex16M",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 127.3899526559214,
+      "24": 127.3899526559214,
+      "28": 127.3899526559214,
+      "32": 127.3899526559214,
+      "36": 127.3899526559214,
+      "40": 127.3899526559214,
+      "44": 127.3899526559214,
+      "48": 127.3899526559214,
+      "52": 127.3899526559214,
+      "56": 127.3899526559214,
+      "60": 127.3899526559214
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/ex16013184",
+      "target_prefix": "",
+      "global_scale": "127.3899526559214",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/ex16013184",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/ex16013184"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..eef17653560e18e42da27248138c38b5af3b6404
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step1500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1500,
+  "tokens_span": 27467776.0,
+  "tokens_sup": 29003776.0,
+  "examples_seen": 1536000,
+  "val_ce": 1.2699265480041504,
+  "val_ce_L20": 1.2315828800201416,
+  "val_ce_L24": 1.2691545486450195,
+  "val_ce_L28": 1.2674776315689087,
+  "val_ce_L32": 1.2868071794509888,
+  "val_ce_L36": 1.1463358402252197,
+  "val_ce_L40": 1.2181566953659058,
+  "val_ce_L44": 1.2027474641799927,
+  "val_ce_L48": 1.2755229473114014,
+  "val_ce_L52": 1.2272130250930786,
+  "val_ce_L56": 1.3846921920776367,
+  "val_ce_L60": 1.4896599054336548,
+  "train_ce": 1.13084731477499,
+  "train_val_gap": 0.13907923322916038,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cex16M",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 127.3899526559214,
+      "24": 127.3899526559214,
+      "28": 127.3899526559214,
+      "32": 127.3899526559214,
+      "36": 127.3899526559214,
+      "40": 127.3899526559214,
+      "44": 127.3899526559214,
+      "48": 127.3899526559214,
+      "52": 127.3899526559214,
+      "56": 127.3899526559214,
+      "60": 127.3899526559214
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/ex16013184",
+      "target_prefix": "",
+      "global_scale": "127.3899526559214",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/ex16013184",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/ex16013184"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..0e3e585613554075ba8498e81fa93747c29bd37e
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2000,
+  "tokens_span": 36853760.0,
+  "tokens_sup": 38901760.0,
+  "examples_seen": 2048000,
+  "val_ce": 1.2440413236618042,
+  "val_ce_L20": 1.1925243139266968,
+  "val_ce_L24": 1.1994342803955078,
+  "val_ce_L28": 1.2157824039459229,
+  "val_ce_L32": 1.2693300247192383,
+  "val_ce_L36": 1.1297534704208374,
+  "val_ce_L40": 1.197664499282837,
+  "val_ce_L44": 1.1829373836517334,
+  "val_ce_L48": 1.2337052822113037,
+  "val_ce_L52": 1.2206411361694336,
+  "val_ce_L56": 1.3781336545944214,
+  "val_ce_L60": 1.489603042602539,
+  "train_ce": 1.0507902012765407,
+  "train_val_gap": 0.19325112238526354,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cex16M",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 127.3899526559214,
+      "24": 127.3899526559214,
+      "28": 127.3899526559214,
+      "32": 127.3899526559214,
+      "36": 127.3899526559214,
+      "40": 127.3899526559214,
+      "44": 127.3899526559214,
+      "48": 127.3899526559214,
+      "52": 127.3899526559214,
+      "56": 127.3899526559214,
+      "60": 127.3899526559214
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/ex16013184",
+      "target_prefix": "",
+      "global_scale": "127.3899526559214",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/ex16013184",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/ex16013184"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..5cb4edb0c87d59de050926c15eca896fe8f93488
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step2500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2500,
+  "tokens_span": 45751296.0,
+  "tokens_sup": 48311296.0,
+  "examples_seen": 2560000,
+  "val_ce": 1.2631806135177612,
+  "val_ce_L20": 1.2181077003479004,
+  "val_ce_L24": 1.235843539237976,
+  "val_ce_L28": 1.2453774213790894,
+  "val_ce_L32": 1.303082823753357,
+  "val_ce_L36": 1.1517521142959595,
+  "val_ce_L40": 1.2191303968429565,
+  "val_ce_L44": 1.1794068813323975,
+  "val_ce_L48": 1.2416424751281738,
+  "val_ce_L52": 1.2120460271835327,
+  "val_ce_L56": 1.399001955986023,
+  "val_ce_L60": 1.5119190216064453,
+  "train_ce": 0.9480317019671202,
+  "train_val_gap": 0.31514891155064106,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cex16M",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 127.3899526559214,
+      "24": 127.3899526559214,
+      "28": 127.3899526559214,
+      "32": 127.3899526559214,
+      "36": 127.3899526559214,
+      "40": 127.3899526559214,
+      "44": 127.3899526559214,
+      "48": 127.3899526559214,
+      "52": 127.3899526559214,
+      "56": 127.3899526559214,
+      "60": 127.3899526559214
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/ex16013184",
+      "target_prefix": "",
+      "global_scale": "127.3899526559214",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/ex16013184",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/ex16013184"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..7257fbcf2de028e6d70df3ce49cab36e7c3e195c
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3000,
+  "tokens_span": 55110656.0,
+  "tokens_sup": 58182656.0,
+  "examples_seen": 3072000,
+  "val_ce": 1.268718957901001,
+  "val_ce_L20": 1.2399290800094604,
+  "val_ce_L24": 1.2517679929733276,
+  "val_ce_L28": 1.225461721420288,
+  "val_ce_L32": 1.3043733835220337,
+  "val_ce_L36": 1.1520649194717407,
+  "val_ce_L40": 1.2319692373275757,
+  "val_ce_L44": 1.1766647100448608,
+  "val_ce_L48": 1.2515302896499634,
+  "val_ce_L52": 1.2295676469802856,
+  "val_ce_L56": 1.4115409851074219,
+  "val_ce_L60": 1.5037416219711304,
+  "train_ce": 0.902420506939292,
+  "train_val_gap": 0.366298450961709,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cex16M",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 127.3899526559214,
+      "24": 127.3899526559214,
+      "28": 127.3899526559214,
+      "32": 127.3899526559214,
+      "36": 127.3899526559214,
+      "40": 127.3899526559214,
+      "44": 127.3899526559214,
+      "48": 127.3899526559214,
+      "52": 127.3899526559214,
+      "56": 127.3899526559214,
+      "60": 127.3899526559214
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/ex16013184",
+      "target_prefix": "",
+      "global_scale": "127.3899526559214",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/ex16013184",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/ex16013184"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..82b91d35da115417915579718ea0fd4563c0d17d
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step3221/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3221,
+  "tokens_span": 58909696.0,
+  "tokens_sup": 62208000.0,
+  "examples_seen": 3298304,
+  "val_ce": 1.2953001260757446,
+  "val_ce_L20": 1.2391730546951294,
+  "val_ce_L24": 1.2797211408615112,
+  "val_ce_L28": 1.26150643825531,
+  "val_ce_L32": 1.3197842836380005,
+  "val_ce_L36": 1.173114538192749,
+  "val_ce_L40": 1.2472410202026367,
+  "val_ce_L44": 1.2259403467178345,
+  "val_ce_L48": 1.2901767492294312,
+  "val_ce_L52": 1.258248209953308,
+  "val_ce_L56": 1.450079083442688,
+  "val_ce_L60": 1.5308386087417603,
+  "train_ce": 0.8167628356301946,
+  "train_val_gap": 0.47853729044555005,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cex16M",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 127.3899526559214,
+      "24": 127.3899526559214,
+      "28": 127.3899526559214,
+      "32": 127.3899526559214,
+      "36": 127.3899526559214,
+      "40": 127.3899526559214,
+      "44": 127.3899526559214,
+      "48": 127.3899526559214,
+      "52": 127.3899526559214,
+      "56": 127.3899526559214,
+      "60": 127.3899526559214
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/ex16013184",
+      "target_prefix": "",
+      "global_scale": "127.3899526559214",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/ex16013184",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/ex16013184"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/lora/README.md b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/lora/adapter_config.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/meta.json b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..fa3ba7fee9be32856db6b9627f93d55d627a53b8
--- /dev/null
+++ b/ab4/inverter/ao.ab4.mean.frombase.s0cex16M/step500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 500,
+  "tokens_span": 9432064.0,
+  "tokens_sup": 9944064.0,
+  "examples_seen": 512000,
+  "val_ce": 1.4110597372055054,
+  "val_ce_L20": 1.4195141792297363,
+  "val_ce_L24": 1.4142688512802124,
+  "val_ce_L28": 1.4495290517807007,
+  "val_ce_L32": 1.4444748163223267,
+  "val_ce_L36": 1.3194388151168823,
+  "val_ce_L40": 1.3338958024978638,
+  "val_ce_L44": 1.3296467065811157,
+  "val_ce_L48": 1.3660451173782349,
+  "val_ce_L52": 1.3605135679244995,
+  "val_ce_L56": 1.5124832391738892,
+  "val_ce_L60": 1.5924752950668335,
+  "train_ce": 1.642555401325226,
+  "train_val_gap": -0.23149566411972056,
+  "config": {
+    "run_name": "ao.ab4.mean.frombase.s0cex16M",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 127.3899526559214,
+      "24": 127.3899526559214,
+      "28": 127.3899526559214,
+      "32": 127.3899526559214,
+      "36": 127.3899526559214,
+      "40": 127.3899526559214,
+      "44": 127.3899526559214,
+      "48": 127.3899526559214,
+      "52": 127.3899526559214,
+      "56": 127.3899526559214,
+      "60": 127.3899526559214
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.mean.s0c/ex16013184",
+      "target_prefix": "",
+      "global_scale": "127.3899526559214",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.mean.s0c/ex16013184",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.mean.s0c/ex16013184"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..1f1a256ae3220bc77b1d02c5829dbb6431efad7d
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1000,
+  "tokens_span": 18284544.0,
+  "tokens_sup": 19308544.0,
+  "examples_seen": 1024000,
+  "val_ce": 1.8243769407272339,
+  "val_ce_L20": 1.8470689058303833,
+  "val_ce_L24": 1.7521394491195679,
+  "val_ce_L28": 1.821311116218567,
+  "val_ce_L32": 1.9243196249008179,
+  "val_ce_L36": 1.7732905149459839,
+  "val_ce_L40": 1.850606918334961,
+  "val_ce_L44": 1.7975739240646362,
+  "val_ce_L48": 1.7537124156951904,
+  "val_ce_L52": 1.7038289308547974,
+  "val_ce_L56": 1.8757108449935913,
+  "val_ce_L60": 1.9609471559524536,
+  "train_ce": 1.9156917350292206,
+  "train_val_gap": -0.09131479430198675,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 500,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..718d62f518ecbb65f0428cd7609f6306dd736d17
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step1500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1500,
+  "tokens_span": 27467776.0,
+  "tokens_sup": 29003776.0,
+  "examples_seen": 1536000,
+  "val_ce": 1.8369470834732056,
+  "val_ce_L20": 1.8444455862045288,
+  "val_ce_L24": 1.730007529258728,
+  "val_ce_L28": 1.8463735580444336,
+  "val_ce_L32": 1.947905421257019,
+  "val_ce_L36": 1.7749509811401367,
+  "val_ce_L40": 1.875942587852478,
+  "val_ce_L44": 1.8229246139526367,
+  "val_ce_L48": 1.770337700843811,
+  "val_ce_L52": 1.7242318391799927,
+  "val_ce_L56": 1.8842061758041382,
+  "val_ce_L60": 1.9778236150741577,
+  "train_ce": 1.780639774799347,
+  "train_val_gap": 0.05630730867385858,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 500,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..5d671f55269ef3ff4273ca0963728c3b3860723f
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2000,
+  "tokens_span": 36853760.0,
+  "tokens_sup": 38901760.0,
+  "examples_seen": 2048000,
+  "val_ce": 1.851524829864502,
+  "val_ce_L20": 1.8631128072738647,
+  "val_ce_L24": 1.7455177307128906,
+  "val_ce_L28": 1.8501631021499634,
+  "val_ce_L32": 1.9608138799667358,
+  "val_ce_L36": 1.7946196794509888,
+  "val_ce_L40": 1.8916082382202148,
+  "val_ce_L44": 1.8425546884536743,
+  "val_ce_L48": 1.7955716848373413,
+  "val_ce_L52": 1.7056642770767212,
+  "val_ce_L56": 1.908909797668457,
+  "val_ce_L60": 2.0017144680023193,
+  "train_ce": 1.674113139629364,
+  "train_val_gap": 0.17741169023513792,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 500,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c3a13736779df8c543a9e1c6db6008a157e7377a
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step2500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2500,
+  "tokens_span": 45751296.0,
+  "tokens_sup": 48311296.0,
+  "examples_seen": 2560000,
+  "val_ce": 1.9243323802947998,
+  "val_ce_L20": 1.9354289770126343,
+  "val_ce_L24": 1.790499210357666,
+  "val_ce_L28": 1.8924976587295532,
+  "val_ce_L32": 2.0343306064605713,
+  "val_ce_L36": 1.8613299131393433,
+  "val_ce_L40": 2.018488645553589,
+  "val_ce_L44": 1.9052304029464722,
+  "val_ce_L48": 1.8771770000457764,
+  "val_ce_L52": 1.763758659362793,
+  "val_ce_L56": 1.9985953569412231,
+  "val_ce_L60": 2.0772664546966553,
+  "train_ce": 1.5780307312011719,
+  "train_val_gap": 0.34630164909362793,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 500,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..2a8571abc8f3cb8516e5e75d56dedd2f2967da4c
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3000,
+  "tokens_span": 55110656.0,
+  "tokens_sup": 58182656.0,
+  "examples_seen": 3072000,
+  "val_ce": 1.9376085996627808,
+  "val_ce_L20": 1.9152202606201172,
+  "val_ce_L24": 1.8374738693237305,
+  "val_ce_L28": 1.8988255262374878,
+  "val_ce_L32": 2.0563125610351562,
+  "val_ce_L36": 1.8899294137954712,
+  "val_ce_L40": 2.0234663486480713,
+  "val_ce_L44": 1.890956997871399,
+  "val_ce_L48": 1.8839918375015259,
+  "val_ce_L52": 1.7914313077926636,
+  "val_ce_L56": 2.0131375789642334,
+  "val_ce_L60": 2.097649097442627,
+  "train_ce": 1.475751217365265,
+  "train_val_gap": 0.46185738229751583,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 500,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..102cac8fb68b53e8f4ac32550d861062ce51223d
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step3221/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3221,
+  "tokens_span": 58909696.0,
+  "tokens_sup": 62208000.0,
+  "examples_seen": 3298304,
+  "val_ce": 1.987030267715454,
+  "val_ce_L20": 1.9688442945480347,
+  "val_ce_L24": 1.8701730966567993,
+  "val_ce_L28": 1.9746431112289429,
+  "val_ce_L32": 2.1030778884887695,
+  "val_ce_L36": 1.9319297075271606,
+  "val_ce_L40": 2.067919969558716,
+  "val_ce_L44": 1.9705668687820435,
+  "val_ce_L48": 1.915732979774475,
+  "val_ce_L52": 1.8143295049667358,
+  "val_ce_L56": 2.0666167736053467,
+  "val_ce_L60": 2.1573173999786377,
+  "train_ce": 1.4055240693674906,
+  "train_val_gap": 0.5815061983479635,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 500,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ae95db514a58b6f27bda4d416d62adc59e8e59a
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend/step500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 500,
+  "tokens_span": 9432064.0,
+  "tokens_sup": 9944064.0,
+  "examples_seen": 512000,
+  "val_ce": 1.913588285446167,
+  "val_ce_L20": 1.9706040620803833,
+  "val_ce_L24": 1.8526281118392944,
+  "val_ce_L28": 1.931719183921814,
+  "val_ce_L32": 2.0084238052368164,
+  "val_ce_L36": 1.8914740085601807,
+  "val_ce_L40": 1.9371665716171265,
+  "val_ce_L44": 1.8915891647338867,
+  "val_ce_L48": 1.8176681995391846,
+  "val_ce_L52": 1.770865797996521,
+  "val_ce_L56": 1.9371360540390015,
+  "val_ce_L60": 2.023968458175659,
+  "train_ce": 2.202586791753769,
+  "train_val_gap": -0.28899850630760193,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..faa36bc353d55fa06da37048da02af8a9f56c00d
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1000,
+  "tokens_span": 18199552.0,
+  "tokens_sup": 19223552.0,
+  "examples_seen": 1024000,
+  "val_ce": 1.7920808792114258,
+  "val_ce_L20": 1.8012419939041138,
+  "val_ce_L24": 1.7134853601455688,
+  "val_ce_L28": 1.787396788597107,
+  "val_ce_L32": 1.894168496131897,
+  "val_ce_L36": 1.7353050708770752,
+  "val_ce_L40": 1.8273862600326538,
+  "val_ce_L44": 1.7786232233047485,
+  "val_ce_L48": 1.7280280590057373,
+  "val_ce_L52": 1.6697708368301392,
+  "val_ce_L56": 1.8291219472885132,
+  "val_ce_L60": 1.9412647485733032,
+  "train_ce": 1.9374957551956178,
+  "train_val_gap": -0.145414875984192,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..e146b6b12ac82bf3749314275f170b29df2cd957
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 10000,
+  "tokens_span": 182770688.0,
+  "tokens_sup": 193010688.0,
+  "examples_seen": 10240000,
+  "val_ce": 1.7749544382095337,
+  "val_ce_L20": 1.7361685037612915,
+  "val_ce_L24": 1.6477519273757935,
+  "val_ce_L28": 1.7213913202285767,
+  "val_ce_L32": 1.8725194931030273,
+  "val_ce_L36": 1.7362204790115356,
+  "val_ce_L40": 1.8407856225967407,
+  "val_ce_L44": 1.7834901809692383,
+  "val_ce_L48": 1.7181828022003174,
+  "val_ce_L52": 1.6599632501602173,
+  "val_ce_L56": 1.8648457527160645,
+  "val_ce_L60": 1.9292012453079224,
+  "train_ce": 1.5870687222480775,
+  "train_val_gap": 0.1878857159614562,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..eee46eb26454396f2f541d26c1855b33b9fbc050
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 10500,
+  "tokens_span": 191866880.0,
+  "tokens_sup": 202618880.0,
+  "examples_seen": 10752000,
+  "val_ce": 1.812445044517517,
+  "val_ce_L20": 1.7721198797225952,
+  "val_ce_L24": 1.6881810426712036,
+  "val_ce_L28": 1.7500368356704712,
+  "val_ce_L32": 1.939782977104187,
+  "val_ce_L36": 1.7865270376205444,
+  "val_ce_L40": 1.8773586750030518,
+  "val_ce_L44": 1.8023074865341187,
+  "val_ce_L48": 1.750296950340271,
+  "val_ce_L52": 1.690260410308838,
+  "val_ce_L56": 1.9031004905700684,
+  "val_ce_L60": 1.960065484046936,
+  "train_ce": 1.5774568073749542,
+  "train_val_gap": 0.2349882371425629,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..a3ebc22b121905e2b1e5576f415f599f8d6f039c
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step10603/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 10603,
+  "tokens_span": 193746944.0,
+  "tokens_sup": 204604416.0,
+  "examples_seen": 10857472,
+  "val_ce": 1.8013019561767578,
+  "val_ce_L20": 1.7613921165466309,
+  "val_ce_L24": 1.680622935295105,
+  "val_ce_L28": 1.7540935277938843,
+  "val_ce_L32": 1.920240879058838,
+  "val_ce_L36": 1.7732092142105103,
+  "val_ce_L40": 1.8595972061157227,
+  "val_ce_L44": 1.8070297241210938,
+  "val_ce_L48": 1.7427419424057007,
+  "val_ce_L52": 1.6692924499511719,
+  "val_ce_L56": 1.8779658079147339,
+  "val_ce_L60": 1.953342318534851,
+  "train_ce": 1.5723324736345161,
+  "train_val_gap": 0.2289694825422417,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..6e76630c347b178a6bfac83d1dbc3cd1b315764b
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step1500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1500,
+  "tokens_span": 27310080.0,
+  "tokens_sup": 28846080.0,
+  "examples_seen": 1536000,
+  "val_ce": 1.7825369834899902,
+  "val_ce_L20": 1.789430022239685,
+  "val_ce_L24": 1.6735433340072632,
+  "val_ce_L28": 1.756948471069336,
+  "val_ce_L32": 1.884681224822998,
+  "val_ce_L36": 1.7312995195388794,
+  "val_ce_L40": 1.816441535949707,
+  "val_ce_L44": 1.7808589935302734,
+  "val_ce_L48": 1.7168846130371094,
+  "val_ce_L52": 1.6651397943496704,
+  "val_ce_L56": 1.8571470975875854,
+  "val_ce_L60": 1.9274777173995972,
+  "train_ce": 1.873959614276886,
+  "train_val_gap": -0.09142263078689572,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..9158e98445c847645e784e363886832d203bba13
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2000,
+  "tokens_span": 36663296.0,
+  "tokens_sup": 38711296.0,
+  "examples_seen": 2048000,
+  "val_ce": 1.7637383937835693,
+  "val_ce_L20": 1.771275520324707,
+  "val_ce_L24": 1.6576613187789917,
+  "val_ce_L28": 1.7454705238342285,
+  "val_ce_L32": 1.871883749961853,
+  "val_ce_L36": 1.6977263689041138,
+  "val_ce_L40": 1.8063582181930542,
+  "val_ce_L44": 1.7593393325805664,
+  "val_ce_L48": 1.7152550220489502,
+  "val_ce_L52": 1.6317721605300903,
+  "val_ce_L56": 1.8167580366134644,
+  "val_ce_L60": 1.923439383506775,
+  "train_ce": 1.8315422201156617,
+  "train_val_gap": -0.06780382633209237,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c0ad65f0b4d37d509c24cf8a8ed4f7cbb5ef338
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step2500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2500,
+  "tokens_span": 45873152.0,
+  "tokens_sup": 48433152.0,
+  "examples_seen": 2560000,
+  "val_ce": 1.773268222808838,
+  "val_ce_L20": 1.7686136960983276,
+  "val_ce_L24": 1.6691566705703735,
+  "val_ce_L28": 1.7607351541519165,
+  "val_ce_L32": 1.8724632263183594,
+  "val_ce_L36": 1.7194207906723022,
+  "val_ce_L40": 1.8170822858810425,
+  "val_ce_L44": 1.7792634963989258,
+  "val_ce_L48": 1.7113780975341797,
+  "val_ce_L52": 1.6428948640823364,
+  "val_ce_L56": 1.8412467241287231,
+  "val_ce_L60": 1.9150534868240356,
+  "train_ce": 1.8090053236484527,
+  "train_val_gap": -0.03573710083961479,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..4a85e8a6f2d5a643e861da5c234e1fd646ff308b
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3000,
+  "tokens_span": 55260160.0,
+  "tokens_sup": 58332160.0,
+  "examples_seen": 3072000,
+  "val_ce": 1.7670562267303467,
+  "val_ce_L20": 1.7637085914611816,
+  "val_ce_L24": 1.6576844453811646,
+  "val_ce_L28": 1.7594772577285767,
+  "val_ce_L32": 1.8642687797546387,
+  "val_ce_L36": 1.7181099653244019,
+  "val_ce_L40": 1.8011683225631714,
+  "val_ce_L44": 1.7788208723068237,
+  "val_ce_L48": 1.7080501317977905,
+  "val_ce_L52": 1.6474641561508179,
+  "val_ce_L56": 1.828676700592041,
+  "val_ce_L60": 1.902827262878418,
+  "train_ce": 1.7642388432025908,
+  "train_val_gap": 0.0028173835277558457,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..2258a9aae5b7c8d80c07194b6cd82ac573f06eff
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step3500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3500,
+  "tokens_span": 64128000.0,
+  "tokens_sup": 67712000.0,
+  "examples_seen": 3584000,
+  "val_ce": 1.7756835222244263,
+  "val_ce_L20": 1.7814521789550781,
+  "val_ce_L24": 1.659296989440918,
+  "val_ce_L28": 1.7426694631576538,
+  "val_ce_L32": 1.8944340944290161,
+  "val_ce_L36": 1.7218436002731323,
+  "val_ce_L40": 1.815621018409729,
+  "val_ce_L44": 1.7770732641220093,
+  "val_ce_L48": 1.7283531427383423,
+  "val_ce_L52": 1.6443930864334106,
+  "val_ce_L56": 1.8534272909164429,
+  "val_ce_L60": 1.9083837270736694,
+  "train_ce": 1.7552919502258302,
+  "train_val_gap": 0.020391571998596092,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..245e89428fe20f5e770df2db837a944c6e26a1c0
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 4000,
+  "tokens_span": 73389056.0,
+  "tokens_sup": 77485056.0,
+  "examples_seen": 4096000,
+  "val_ce": 1.7571723461151123,
+  "val_ce_L20": 1.7551218271255493,
+  "val_ce_L24": 1.6319456100463867,
+  "val_ce_L28": 1.7203583717346191,
+  "val_ce_L32": 1.8491640090942383,
+  "val_ce_L36": 1.720122218132019,
+  "val_ce_L40": 1.7891980409622192,
+  "val_ce_L44": 1.7556484937667847,
+  "val_ce_L48": 1.70125412940979,
+  "val_ce_L52": 1.640007495880127,
+  "val_ce_L56": 1.8401392698287964,
+  "val_ce_L60": 1.9174529314041138,
+  "train_ce": 1.7272225964069365,
+  "train_val_gap": 0.029949749708175766,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..da3885ddc97953c6903168030fa493f381a4ece9
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step4500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 4500,
+  "tokens_span": 82491392.0,
+  "tokens_sup": 87099392.0,
+  "examples_seen": 4608000,
+  "val_ce": 1.7568258047103882,
+  "val_ce_L20": 1.7627531290054321,
+  "val_ce_L24": 1.641499400138855,
+  "val_ce_L28": 1.7271894216537476,
+  "val_ce_L32": 1.8694969415664673,
+  "val_ce_L36": 1.719331979751587,
+  "val_ce_L40": 1.7741148471832275,
+  "val_ce_L44": 1.755820393562317,
+  "val_ce_L48": 1.7077573537826538,
+  "val_ce_L52": 1.6280866861343384,
+  "val_ce_L56": 1.8304780721664429,
+  "val_ce_L60": 1.903746485710144,
+  "train_ce": 1.7185818226337433,
+  "train_val_gap": 0.03824398207664492,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c115fd137ad3530f551fea7834c9c03feba6cd4
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 500,
+  "tokens_span": 8822784.0,
+  "tokens_sup": 9334784.0,
+  "examples_seen": 512000,
+  "val_ce": 1.8743270635604858,
+  "val_ce_L20": 1.9060173034667969,
+  "val_ce_L24": 1.7950860261917114,
+  "val_ce_L28": 1.8753223419189453,
+  "val_ce_L32": 1.9952141046524048,
+  "val_ce_L36": 1.816349744796753,
+  "val_ce_L40": 1.8923569917678833,
+  "val_ce_L44": 1.8534454107284546,
+  "val_ce_L48": 1.8056446313858032,
+  "val_ce_L52": 1.744409441947937,
+  "val_ce_L56": 1.9133418798446655,
+  "val_ce_L60": 2.0156219005584717,
+  "train_ce": 2.216562216758728,
+  "train_val_gap": -0.34223515319824216,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..d3789683880ac66814855280c4a45a33ef1bb435
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 5000,
+  "tokens_span": 91223040.0,
+  "tokens_sup": 96343040.0,
+  "examples_seen": 5120000,
+  "val_ce": 1.7648180723190308,
+  "val_ce_L20": 1.754724144935608,
+  "val_ce_L24": 1.650732398033142,
+  "val_ce_L28": 1.721174716949463,
+  "val_ce_L32": 1.8543258905410767,
+  "val_ce_L36": 1.7271512746810913,
+  "val_ce_L40": 1.796573281288147,
+  "val_ce_L44": 1.7782477140426636,
+  "val_ce_L48": 1.7166303396224976,
+  "val_ce_L52": 1.6516008377075195,
+  "val_ce_L56": 1.8516117334365845,
+  "val_ce_L60": 1.9031791687011719,
+  "train_ce": 1.7014448640346527,
+  "train_val_gap": 0.06337320828437809,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..953ff1af6363db1ef227513a9b97156eb1d3233b
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step5500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 5500,
+  "tokens_span": 100399104.0,
+  "tokens_sup": 106031104.0,
+  "examples_seen": 5632000,
+  "val_ce": 1.7886992692947388,
+  "val_ce_L20": 1.7729696035385132,
+  "val_ce_L24": 1.6577876806259155,
+  "val_ce_L28": 1.7623907327651978,
+  "val_ce_L32": 1.8782734870910645,
+  "val_ce_L36": 1.7539805173873901,
+  "val_ce_L40": 1.8335860967636108,
+  "val_ce_L44": 1.808057427406311,
+  "val_ce_L48": 1.7392925024032593,
+  "val_ce_L52": 1.6602534055709839,
+  "val_ce_L56": 1.8757120370864868,
+  "val_ce_L60": 1.923459529876709,
+  "train_ce": 1.6714441981315613,
+  "train_val_gap": 0.11725507116317746,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..47a035b013040b45dea366b8530988761306f8db
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 6000,
+  "tokens_span": 109400064.0,
+  "tokens_sup": 115544064.0,
+  "examples_seen": 6144000,
+  "val_ce": 1.764783501625061,
+  "val_ce_L20": 1.773260474205017,
+  "val_ce_L24": 1.6444019079208374,
+  "val_ce_L28": 1.7169133424758911,
+  "val_ce_L32": 1.8760501146316528,
+  "val_ce_L36": 1.7275444269180298,
+  "val_ce_L40": 1.8197300434112549,
+  "val_ce_L44": 1.7778891324996948,
+  "val_ce_L48": 1.6996904611587524,
+  "val_ce_L52": 1.63585364818573,
+  "val_ce_L56": 1.8417768478393555,
+  "val_ce_L60": 1.8857070207595825,
+  "train_ce": 1.6733876857757568,
+  "train_val_gap": 0.09139581584930423,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..1341b1a904ab1cdce55fddf443adcb476e67875c
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step6500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 6500,
+  "tokens_span": 118416384.0,
+  "tokens_sup": 125072384.0,
+  "examples_seen": 6656000,
+  "val_ce": 1.7873486280441284,
+  "val_ce_L20": 1.7812312841415405,
+  "val_ce_L24": 1.6542712450027466,
+  "val_ce_L28": 1.7637887001037598,
+  "val_ce_L32": 1.8974567651748657,
+  "val_ce_L36": 1.749007225036621,
+  "val_ce_L40": 1.8091329336166382,
+  "val_ce_L44": 1.8173317909240723,
+  "val_ce_L48": 1.7302439212799072,
+  "val_ce_L52": 1.6731115579605103,
+  "val_ce_L56": 1.8600635528564453,
+  "val_ce_L60": 1.9184402227401733,
+  "train_ce": 1.6520842709541321,
+  "train_val_gap": 0.1352643570899963,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..d927973cde4a1b240211b1e80d57a74866ec3656
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 7000,
+  "tokens_span": 127619072.0,
+  "tokens_sup": 134787072.0,
+  "examples_seen": 7168000,
+  "val_ce": 1.8223226070404053,
+  "val_ce_L20": 1.793555736541748,
+  "val_ce_L24": 1.654773235321045,
+  "val_ce_L28": 1.7898865938186646,
+  "val_ce_L32": 1.9336228370666504,
+  "val_ce_L36": 1.7956457138061523,
+  "val_ce_L40": 1.873663067817688,
+  "val_ce_L44": 1.8633251190185547,
+  "val_ce_L48": 1.7706199884414673,
+  "val_ce_L52": 1.7119966745376587,
+  "val_ce_L56": 1.8956133127212524,
+  "val_ce_L60": 1.950117588043213,
+  "train_ce": 1.6343621151447296,
+  "train_val_gap": 0.18796049189567565,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..c1f2f82af2de79ec675532db84869f5f66a27856
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step7500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 7500,
+  "tokens_span": 136687616.0,
+  "tokens_sup": 144367616.0,
+  "examples_seen": 7680000,
+  "val_ce": 1.806792140007019,
+  "val_ce_L20": 1.782148838043213,
+  "val_ce_L24": 1.6564226150512695,
+  "val_ce_L28": 1.767816185951233,
+  "val_ce_L32": 1.934617519378662,
+  "val_ce_L36": 1.7793848514556885,
+  "val_ce_L40": 1.8454653024673462,
+  "val_ce_L44": 1.813987374305725,
+  "val_ce_L48": 1.7581870555877686,
+  "val_ce_L52": 1.7106761932373047,
+  "val_ce_L56": 1.8819078207015991,
+  "val_ce_L60": 1.934120774269104,
+  "train_ce": 1.628008695602417,
+  "train_val_gap": 0.17878344440460214,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..1e77a010a32f2555c898dfd2d93455621ea0ba57
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 8000,
+  "tokens_span": 145676288.0,
+  "tokens_sup": 153868288.0,
+  "examples_seen": 8192000,
+  "val_ce": 1.80190110206604,
+  "val_ce_L20": 1.755014419555664,
+  "val_ce_L24": 1.6519068479537964,
+  "val_ce_L28": 1.7656720876693726,
+  "val_ce_L32": 1.9143422842025757,
+  "val_ce_L36": 1.7774903774261475,
+  "val_ce_L40": 1.8421233892440796,
+  "val_ce_L44": 1.835668921470642,
+  "val_ce_L48": 1.745100736618042,
+  "val_ce_L52": 1.7084640264511108,
+  "val_ce_L56": 1.8772773742675781,
+  "val_ce_L60": 1.9357494115829468,
+  "train_ce": 1.615908712863922,
+  "train_val_gap": 0.18599238920211802,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..aa46ec50e2f2f60a7a9f8df494db480ba35fdde4
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step8500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 8500,
+  "tokens_span": 154830848.0,
+  "tokens_sup": 163534848.0,
+  "examples_seen": 8704000,
+  "val_ce": 1.831396222114563,
+  "val_ce_L20": 1.7959104776382446,
+  "val_ce_L24": 1.6802988052368164,
+  "val_ce_L28": 1.7868832349777222,
+  "val_ce_L32": 1.9598945379257202,
+  "val_ce_L36": 1.7931627035140991,
+  "val_ce_L40": 1.8816440105438232,
+  "val_ce_L44": 1.8572529554367065,
+  "val_ce_L48": 1.7821193933486938,
+  "val_ce_L52": 1.7318612337112427,
+  "val_ce_L56": 1.9063631296157837,
+  "val_ce_L60": 1.9597517251968384,
+  "train_ce": 1.605482304573059,
+  "train_val_gap": 0.22591391754150392,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..35dd7cda01e3b61dacb860b1b2123cde703185c3
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 9000,
+  "tokens_span": 164231168.0,
+  "tokens_sup": 173447168.0,
+  "examples_seen": 9216000,
+  "val_ce": 1.7828609943389893,
+  "val_ce_L20": 1.738173007965088,
+  "val_ce_L24": 1.6738163232803345,
+  "val_ce_L28": 1.740577220916748,
+  "val_ce_L32": 1.8968254327774048,
+  "val_ce_L36": 1.7528197765350342,
+  "val_ce_L40": 1.8340978622436523,
+  "val_ce_L44": 1.781446099281311,
+  "val_ce_L48": 1.7329750061035156,
+  "val_ce_L52": 1.6805695295333862,
+  "val_ce_L56": 1.8642674684524536,
+  "val_ce_L60": 1.9040536880493164,
+  "train_ce": 1.5953492527008057,
+  "train_val_gap": 0.18751174163818352,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..685aa6c564b57c0751f4dd69daa4cd41ee6ea452
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.e1cend_full/step9500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 9500,
+  "tokens_span": 173485056.0,
+  "tokens_sup": 183213056.0,
+  "examples_seen": 9728000,
+  "val_ce": 1.8191126585006714,
+  "val_ce_L20": 1.777260184288025,
+  "val_ce_L24": 1.6863274574279785,
+  "val_ce_L28": 1.7733235359191895,
+  "val_ce_L32": 1.9419864416122437,
+  "val_ce_L36": 1.7941138744354248,
+  "val_ce_L40": 1.8848170042037964,
+  "val_ce_L44": 1.8223129510879517,
+  "val_ce_L48": 1.7580633163452148,
+  "val_ce_L52": 1.7012125253677368,
+  "val_ce_L56": 1.902970314025879,
+  "val_ce_L60": 1.950891137123108,
+  "train_ce": 1.5933048388957978,
+  "train_val_gap": 0.22580781960487362,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.e1cend_full",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 105.20583130160833,
+      "24": 105.20583130160833,
+      "28": 105.20583130160833,
+      "32": 105.20583130160833,
+      "36": 105.20583130160833,
+      "40": 105.20583130160833,
+      "44": 105.20583130160833,
+      "48": 105.20583130160833,
+      "52": 105.20583130160833,
+      "56": 105.20583130160833,
+      "60": 105.20583130160833
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 10891600,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.e1c/end",
+      "target_prefix": "",
+      "global_scale": "105.20583130160833",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2_n9579668/ar.ab4.rand.e1c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.e1c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..11118e2aac209adfe5ac55e832501d9684d4b639
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1000,
+  "tokens_span": 18284544.0,
+  "tokens_sup": 19308544.0,
+  "examples_seen": 1024000,
+  "val_ce": 1.81086266040802,
+  "val_ce_L20": 1.8411787748336792,
+  "val_ce_L24": 1.7450300455093384,
+  "val_ce_L28": 1.8085185289382935,
+  "val_ce_L32": 1.890278697013855,
+  "val_ce_L36": 1.7705267667770386,
+  "val_ce_L40": 1.8333673477172852,
+  "val_ce_L44": 1.7766079902648926,
+  "val_ce_L48": 1.7185347080230713,
+  "val_ce_L52": 1.693217158317566,
+  "val_ce_L56": 1.8723613023757935,
+  "val_ce_L60": 1.9574521780014038,
+  "train_ce": 1.8991861789226532,
+  "train_val_gap": -0.08832351851463316,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 112.6079551190218,
+      "24": 112.6079551190218,
+      "28": 112.6079551190218,
+      "32": 112.6079551190218,
+      "36": 112.6079551190218,
+      "40": 112.6079551190218,
+      "44": 112.6079551190218,
+      "48": 112.6079551190218,
+      "52": 112.6079551190218,
+      "56": 112.6079551190218,
+      "60": 112.6079551190218
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.s0c/end",
+      "target_prefix": "",
+      "global_scale": "112.6079551190218",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..ed7d666c7993f4fb2734902e14fdaed701390dd7
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step1500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 1500,
+  "tokens_span": 27467776.0,
+  "tokens_sup": 29003776.0,
+  "examples_seen": 1536000,
+  "val_ce": 1.816232681274414,
+  "val_ce_L20": 1.840232253074646,
+  "val_ce_L24": 1.7090708017349243,
+  "val_ce_L28": 1.8165968656539917,
+  "val_ce_L32": 1.907020926475525,
+  "val_ce_L36": 1.7790488004684448,
+  "val_ce_L40": 1.8538459539413452,
+  "val_ce_L44": 1.7911745309829712,
+  "val_ce_L48": 1.728451132774353,
+  "val_ce_L52": 1.6967592239379883,
+  "val_ce_L56": 1.8687410354614258,
+  "val_ce_L60": 1.972917079925537,
+  "train_ce": 1.760792727947235,
+  "train_val_gap": 0.05543995332717899,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 112.6079551190218,
+      "24": 112.6079551190218,
+      "28": 112.6079551190218,
+      "32": 112.6079551190218,
+      "36": 112.6079551190218,
+      "40": 112.6079551190218,
+      "44": 112.6079551190218,
+      "48": 112.6079551190218,
+      "52": 112.6079551190218,
+      "56": 112.6079551190218,
+      "60": 112.6079551190218
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.s0c/end",
+      "target_prefix": "",
+      "global_scale": "112.6079551190218",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..43007bab5dbd471fe1bd44a2e98db0b7e804677d
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2000,
+  "tokens_span": 36853760.0,
+  "tokens_sup": 38901760.0,
+  "examples_seen": 2048000,
+  "val_ce": 1.841150164604187,
+  "val_ce_L20": 1.8522392511367798,
+  "val_ce_L24": 1.7343697547912598,
+  "val_ce_L28": 1.821567416191101,
+  "val_ce_L32": 1.9364312887191772,
+  "val_ce_L36": 1.803939938545227,
+  "val_ce_L40": 1.874720573425293,
+  "val_ce_L44": 1.811169981956482,
+  "val_ce_L48": 1.7826001644134521,
+  "val_ce_L52": 1.704633355140686,
+  "val_ce_L56": 1.9110597372055054,
+  "val_ce_L60": 2.0107710361480713,
+  "train_ce": 1.6551705014705658,
+  "train_val_gap": 0.18597966313362124,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 112.6079551190218,
+      "24": 112.6079551190218,
+      "28": 112.6079551190218,
+      "32": 112.6079551190218,
+      "36": 112.6079551190218,
+      "40": 112.6079551190218,
+      "44": 112.6079551190218,
+      "48": 112.6079551190218,
+      "52": 112.6079551190218,
+      "56": 112.6079551190218,
+      "60": 112.6079551190218
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.s0c/end",
+      "target_prefix": "",
+      "global_scale": "112.6079551190218",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/lora/adapter_config.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/lora/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..003eac4a729e52731cbbeb7b212cd98a0a9e2a91
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/lora/adapter_config.json
@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Qwen3_5ForCausalLM",
+    "parent_library": "transformers.models.qwen3_5.modeling_qwen3_5"
+  },
+  "base_model_name_or_path": "Qwen/Qwen3.6-27B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "up_proj",
+    "o_proj",
+    "in_proj_qkv",
+    "q_proj",
+    "v_proj",
+    "out_proj",
+    "in_proj_z",
+    "k_proj",
+    "down_proj",
+    "in_proj_b",
+    "in_proj_a",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..73b9269bf85a1d355308400952d50a0a99325057
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step2500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 2500,
+  "tokens_span": 45751296.0,
+  "tokens_sup": 48311296.0,
+  "examples_seen": 2560000,
+  "val_ce": 1.9162784814834595,
+  "val_ce_L20": 1.9151865243911743,
+  "val_ce_L24": 1.8112930059432983,
+  "val_ce_L28": 1.8760007619857788,
+  "val_ce_L32": 2.014254570007324,
+  "val_ce_L36": 1.8520971536636353,
+  "val_ce_L40": 1.9863215684890747,
+  "val_ce_L44": 1.8757704496383667,
+  "val_ce_L48": 1.8617194890975952,
+  "val_ce_L52": 1.7628506422042847,
+  "val_ce_L56": 1.9889789819717407,
+  "val_ce_L60": 2.1245224475860596,
+  "train_ce": 1.5593324184417725,
+  "train_val_gap": 0.35694606304168697,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 112.6079551190218,
+      "24": 112.6079551190218,
+      "28": 112.6079551190218,
+      "32": 112.6079551190218,
+      "36": 112.6079551190218,
+      "40": 112.6079551190218,
+      "44": 112.6079551190218,
+      "48": 112.6079551190218,
+      "52": 112.6079551190218,
+      "56": 112.6079551190218,
+      "60": 112.6079551190218
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.s0c/end",
+      "target_prefix": "",
+      "global_scale": "112.6079551190218",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3000/lora/README.md b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3000/lora/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..0e88429bac1e7dfe0c7a41603b1c9f71ff730e85
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3000/lora/README.md
@@ -0,0 +1,206 @@
+---
+base_model: Qwen/Qwen3.6-27B
+library_name: peft
+tags:
+- base_model:adapter:Qwen/Qwen3.6-27B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3000/meta.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3000/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c1a0cd65b1b983c63f66a5062fa64253aa26219
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3000/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3000,
+  "tokens_span": 55110656.0,
+  "tokens_sup": 58182656.0,
+  "examples_seen": 3072000,
+  "val_ce": 1.9286749362945557,
+  "val_ce_L20": 1.9278453588485718,
+  "val_ce_L24": 1.836302399635315,
+  "val_ce_L28": 1.8771061897277832,
+  "val_ce_L32": 2.027836561203003,
+  "val_ce_L36": 1.8890460729599,
+  "val_ce_L40": 1.9724657535552979,
+  "val_ce_L44": 1.882314682006836,
+  "val_ce_L48": 1.8802251815795898,
+  "val_ce_L52": 1.7948040962219238,
+  "val_ce_L56": 2.0068376064300537,
+  "val_ce_L60": 2.1118714809417725,
+  "train_ce": 1.4606372919082642,
+  "train_val_gap": 0.4680376443862915,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 112.6079551190218,
+      "24": 112.6079551190218,
+      "28": 112.6079551190218,
+      "32": 112.6079551190218,
+      "36": 112.6079551190218,
+      "40": 112.6079551190218,
+      "44": 112.6079551190218,
+      "48": 112.6079551190218,
+      "52": 112.6079551190218,
+      "56": 112.6079551190218,
+      "60": 112.6079551190218
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.s0c/end",
+      "target_prefix": "",
+      "global_scale": "112.6079551190218",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3221/meta.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3221/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..9302269ffd071c59fa4aaae1ab41ce3ee57a961d
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step3221/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 3221,
+  "tokens_span": 58909696.0,
+  "tokens_sup": 62208000.0,
+  "examples_seen": 3298304,
+  "val_ce": 1.9684505462646484,
+  "val_ce_L20": 1.9270482063293457,
+  "val_ce_L24": 1.8800283670425415,
+  "val_ce_L28": 1.9175580739974976,
+  "val_ce_L32": 2.0880651473999023,
+  "val_ce_L36": 1.9327256679534912,
+  "val_ce_L40": 1.9936017990112305,
+  "val_ce_L44": 1.959527611732483,
+  "val_ce_L48": 1.902092695236206,
+  "val_ce_L52": 1.8203984498977661,
+  "val_ce_L56": 2.0570197105407715,
+  "val_ce_L60": 2.1655914783477783,
+  "train_ce": 1.3880018756400405,
+  "train_val_gap": 0.5804486706246079,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 112.6079551190218,
+      "24": 112.6079551190218,
+      "28": 112.6079551190218,
+      "32": 112.6079551190218,
+      "36": 112.6079551190218,
+      "40": 112.6079551190218,
+      "44": 112.6079551190218,
+      "48": 112.6079551190218,
+      "52": 112.6079551190218,
+      "56": 112.6079551190218,
+      "60": 112.6079551190218
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.s0c/end",
+      "target_prefix": "",
+      "global_scale": "112.6079551190218",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file
diff --git a/ab4/inverter/ao.ab4.rand.frombase.s0cend/step500/meta.json b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step500/meta.json
new file mode 100644
index 0000000000000000000000000000000000000000..3380d3089371b5bc7b2b98077415d38061283fa1
--- /dev/null
+++ b/ab4/inverter/ao.ab4.rand.frombase.s0cend/step500/meta.json
@@ -0,0 +1,104 @@
+{
+  "micro_steps": 500,
+  "tokens_span": 9432064.0,
+  "tokens_sup": 9944064.0,
+  "examples_seen": 512000,
+  "val_ce": 1.8858362436294556,
+  "val_ce_L20": 1.9395192861557007,
+  "val_ce_L24": 1.8277157545089722,
+  "val_ce_L28": 1.9054198265075684,
+  "val_ce_L32": 1.968565821647644,
+  "val_ce_L36": 1.8748693466186523,
+  "val_ce_L40": 1.9044348001480103,
+  "val_ce_L44": 1.863782286643982,
+  "val_ce_L48": 1.772801160812378,
+  "val_ce_L52": 1.7449232339859009,
+  "val_ce_L56": 1.9120397567749023,
+  "val_ce_L60": 2.0100181102752686,
+  "train_ce": 2.1935681476593016,
+  "train_val_gap": -0.30773190402984607,
+  "config": {
+    "run_name": "ao.ab4.rand.frombase.s0cend",
+    "transform": "scaled",
+    "layers": [
+      20,
+      24,
+      28,
+      32,
+      36,
+      40,
+      44,
+      48,
+      52,
+      56,
+      60
+    ],
+    "alpha": 16000.0,
+    "clip_mult": 4.0,
+    "scales": {
+      "20": 112.6079551190218,
+      "24": 112.6079551190218,
+      "28": 112.6079551190218,
+      "32": 112.6079551190218,
+      "36": 112.6079551190218,
+      "40": 112.6079551190218,
+      "44": 112.6079551190218,
+      "48": 112.6079551190218,
+      "52": 112.6079551190218,
+      "56": 112.6079551190218,
+      "60": 112.6079551190218
+    },
+    "freeze_base": false,
+    "xform_hidden": 0,
+    "xform_arch": "resmlp",
+    "logits_on_targets": false,
+    "prefix_cache": false,
+    "n_examples": 3333472,
+    "min_len": 1,
+    "max_len": 64,
+    "lora_r": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "lr": 0.0003,
+    "lr_sched": "constant",
+    "micro_batch": 128,
+    "val_micro_batch": 16,
+    "grad_accum": 8,
+    "epochs": 1,
+    "eval_every_steps": 500,
+    "max_eval_rows": 1024,
+    "warmup_steps": 100,
+    "seed": 0,
+    "init_from": "",
+    "skip_examples": 0,
+    "skip_batches": 0,
+    "exclude_idx_file": "",
+    "num_workers": 2,
+    "extra": {
+      "kind": "ao_ladder",
+      "ar_run": "ar.ab4.rand.s0c/end",
+      "target_prefix": "",
+      "global_scale": "112.6079551190218",
+      "pool": "ao_pool/pool_iolens_u64w2.safetensors",
+      "scale_mode": "global_frozen",
+      "grad_ckpt": "on",
+      "compile_blocks": "off",
+      "prompt": "continuation_raw",
+      "layers_per_crop": "4",
+      "val_source": "pool",
+      "max_len_filter": "32",
+      "adopt_arout_picks": "False",
+      "split_seed": "1234",
+      "arout_dir": "/scratch/ola/ao_arout_u64w2/ar.ab4.rand.s0c/end",
+      "val_pool": "ao_pool/pool_iolens_u64.safetensors",
+      "val_arout_dir": "ao_arout_u64/ar.ab4.rand.s0c/end"
+    },
+    "exact_resume": false,
+    "decay_frac": 0.1,
+    "warmup_examples": 0,
+    "expected_effective_batch": 0,
+    "eval_every_examples": 0,
+    "train_examples_cap": 0,
+    "total_opt_steps_override": null
+  }
+}
\ No newline at end of file