OpenCodifier Decision 2B v6full arm (t33p): gated release, PASS x2

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+t33p-2b-v6full-lora-F16.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,75 @@
+---
+license: apache-2.0
+base_model: Qwen/Qwen3.5-2B
+library_name: peft
+tags:
+- opencodifier
+- decision-model
+- lora
+- structured-outputs
+- qwen3_5
+---
+
+# OpenCodifier Decision 2B — v6full arm (run t33p)
+
+A LoRA decision head for [Qwen/Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B)
+trained by the **OpenCodifier** decision runtime (Apache-2.0). The model maps
+unstructured state to small, machine-actionable typed decisions — **choice**
+(one of a runtime-defined candidate list), **boolean** (true/false with
+probability), and **score** (ordered levels) — with calibrated confidence and
+deterministic abstention. It is not a chat model; free-form generation fields
+are unsupported by design (`unsupported_generation_field`).
+
+## Numbers (one number per surface, never a universal claim)
+
+| surface | n | accuracy | note |
+|---|---|---|---|
+| OpenCodifier suite holdout | 120 | **0.6917** | `oc-readout-v1` readout, fp16 |
+| suite relational slice (rel200) | 200 | **0.6607** | same readout |
+| JevBench public split | 231 | **0.6580** | family-macro, official harness `fstandhartinger/jevbench` @ `9ec6f15a`, llama.cpp fork lane (Q4_K_M base + F16 LoRA) |
+
+**Release gate (suite ≥ 0.68 AND JevBench ≥ 0.6494): PASS × 2.**
+This arm's JevBench row beats its base and the community 0.8B checkpoint row
+(0.6494) on both surfaces.
+
+Calibration: raw winner probabilities are overconfident (true everywhere we
+measured). Ship confidence only through the fitted per-kind temperature —
+`perkind_calibration.json` (T = 5.225, post-fit ECE 0.1313 on the holdout)
+ships with this repo. Raw softmax is not calibrated confidence.
+
+## Training
+
+- Base: `Qwen/Qwen3.5-2B` (Apache-2.0), LoRA r=16 / α=32 / dropout 0.05
+- 1 epoch, lr 1e-4 cosine, warmup 3%, effective batch 16 (2×8 accum),
+  max_len 2048, fp16 + sdpa, seed 42 → 19,058 steps (~9.7 h, single GPU)
+- Data: withheld decision corpus, 306,999 rows (151,436 choice / 81,537
+  score / 72,026 no-outcome), confidence-gated teacher labels with
+  two-family consensus lineage (Apache-2.0 / CC0 sources only)
+- Full recipe in `train-manifest.json`
+
+## What is withheld, and why
+
+The training corpus, the teacher-facts, and the extraction/consensus
+machinery (VIVERE lane) are **withheld**. Weights alone are inspectable and
+replicable with a comparable corpus; the corpus — multi-family consensus over
+183k+ typed decision labels with a per-question disputes ledger — is the part
+that is not cheaply reproducible by a single-model distill. This is the
+"Claims we refuse to make" posture: no fully-open claims, the withholding is
+stated plainly.
+
+## Usage
+
+- **llama.cpp**: `llama-server -m Qwen3.5-2B-Q4_K_M.gguf --lora t33p-2b-v6full-lora-F16.gguf`
+  (the board row above was produced through exactly this lane)
+- **transformers/PEFT**: load `Qwen/Qwen3.5-2B` + `PeftModel.from_pretrained`
+  with this adapter
+- Download is opt-in: the OpenCodifier engine never auto-pulls weights at
+  build or run time.
+
+## Engine
+
+The engine that serves, verifies, and calibrates this model — rules,
+relational solver, BM25 narrowing, candidate-conditioned decision math in
+Rust f64, confidence-gated verification — is Apache-2.0 at
+[OpenCodifier](https://github.com/aliasfoxkde/OpenCodifier) (docs:
+`docs/BENCHMARKS.md` is the comparison page of record).

adapter-best/adapter_config.json

@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "/home/mkinney/oc-model-eval/models/hf/Qwen3.5-2B",
+  "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,
+  "kasa_config": null,
+  "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,
+  "monteclora_config": null,
+  "peft_type": "LORA",
+  "peft_version": "0.21.0",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "o_proj",
+    "k_proj",
+    "q_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false,
+  "velora_config": null
+}

adapter-best/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00b9c542ad8ee6cef322316bf85f0dbc8e788b6b567d53f1c6dbfd6cd80c7fe8
+size 5904616

adapter-best/perkind_calibration.json

@@ -0,0 +1,11 @@
+{
+  "acc": 0.6916666666666667,
+  "ece_all_perk": 0.13132679659121776,
+  "ece_per_kind": {
+    "choice": 0.13132679659121776
+  },
+  "per_kind_T": {
+    "choice": 5.225350451126656
+  },
+  "rows": 120
+}

t33p-2b-v6full-lora-F16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14971730dbc0105940f823645ac283fe06852b278bd76dec281adb62f77ef0fa
+size 2953088

train-manifest.json

@@ -0,0 +1,549 @@
+{
+ "manifest_version": "opencodifier.e1-train/5",
+ "resume": null,
+ "ckpt_every": 100,
+ "loss_mode": "hard-verdict-ce",
+ "base": "/home/mkinney/oc-model-eval/models/hf/Qwen3.5-2B",
+ "data": {
+  "path": "/home/mkinney/oc-model-eval/corpora/merged-v6-full/e1-sft-v1.rows.jsonl",
+  "sha256": "d4aeca494955fa1531ccc3681af0763e6c21a1e7b4e2ff6d02e698033c3331df",
+  "rows": 306999,
+  "train": 304936,
+  "val": 1999,
+  "loud_skipped_over_maxlen_train": 63,
+  "loud_skipped_over_maxlen_val": 1,
+  "soft_bad_slotprobs_train": 0,
+  "soft_bad_slotprobs_val": 0
+ },
+ "row_weight": {
+  "field": "",
+  "train_rows": 304936,
+  "train_nonunit": 0,
+  "train_mean": 1.0,
+  "val_nonunit": 0
+ },
+ "knobs": {
+  "epochs": 1.0,
+  "lr": 0.0001,
+  "lora_r": 16,
+  "lora_alpha": 32,
+  "lora_dropout": 0.05,
+  "target_modules": [
+   "q_proj",
+   "k_proj",
+   "v_proj",
+   "o_proj"
+  ],
+  "batch": 2,
+  "accum": 8,
+  "max_len": 2048,
+  "dtype": "fp16",
+  "attn": "sdpa",
+  "warmup": 0.03,
+  "seed": 42,
+  "torch_seeded": true,
+  "val_frac": 0.01,
+  "val_cap": 2000,
+  "val_every": 500,
+  "family_weight": {},
+  "length_bucket": false,
+  "grad_checkpoint": true,
+  "soft_labels": false,
+  "teacher_temp": 1.0,
+  "stop_weight": 1.0
+ },
+ "optimizer": {
+  "name": "AdamW",
+  "weight_decay": 0.01,
+  "betas": [
+   0.9,
+   0.999
+  ],
+  "eps": 1e-08
+ },
+ "versions": {
+  "python": "3.12.15",
+  "torch": "2.14.0+cu130",
+  "transformers": "5.17.0",
+  "peft": "0.21.0"
+ },
+ "steps": 19058,
+ "final_loss_token_weighted": 0.0648,
+ "tokens_per_second_wall": 44,
+ "peak_memory_bytes": 4989219328,
+ "loss_surface": "verdict tokens only (y=1 segments + EOS)",
+ "val_history": [
+  {
+   "step": 500,
+   "loss": 0.1197,
+   "tok_acc": 0.952,
+   "entropy": 0.0788,
+   "per_family": {
+    "choice": 0.1605,
+    "noul": 0.0704,
+    "relational_compositional": 0.0171,
+    "score": 0.1543
+   }
+  },
+  {
+   "step": 1000,
+   "loss": 0.0912,
+   "tok_acc": 0.9624,
+   "entropy": 0.0972,
+   "per_family": {
+    "choice": 0.1188,
+    "noul": 0.0587,
+    "relational_compositional": 0.0102,
+    "score": 0.1209
+   }
+  },
+  {
+   "step": 1500,
+   "loss": 0.0808,
+   "tok_acc": 0.9652,
+   "entropy": 0.0745,
+   "per_family": {
+    "choice": 0.1158,
+    "noul": 0.0652,
+    "relational_compositional": 0.0036,
+    "score": 0.0915
+   }
+  },
+  {
+   "step": 2000,
+   "loss": 0.0901,
+   "tok_acc": 0.9644,
+   "entropy": 0.0551,
+   "per_family": {
+    "choice": 0.1182,
+    "noul": 0.0577,
+    "relational_compositional": 0.0098,
+    "score": 0.1189
+   }
+  },
+  {
+   "step": 2500,
+   "loss": 0.0716,
+   "tok_acc": 0.9702,
+   "entropy": 0.0695,
+   "per_family": {
+    "choice": 0.1068,
+    "noul": 0.0509,
+    "relational_compositional": 0.0031,
+    "score": 0.0797
+   }
+  },
+  {
+   "step": 3000,
+   "loss": 0.0655,
+   "tok_acc": 0.9742,
+   "entropy": 0.0522,
+   "per_family": {
+    "choice": 0.0959,
+    "noul": 0.0372,
+    "relational_compositional": 0.0019,
+    "score": 0.0797
+   }
+  },
+  {
+   "step": 3500,
+   "loss": 0.0613,
+   "tok_acc": 0.975,
+   "entropy": 0.0545,
+   "per_family": {
+    "choice": 0.0862,
+    "noul": 0.046,
+    "relational_compositional": 0.0036,
+    "score": 0.0726
+   }
+  },
+  {
+   "step": 4000,
+   "loss": 0.0644,
+   "tok_acc": 0.9734,
+   "entropy": 0.0569,
+   "per_family": {
+    "choice": 0.0908,
+    "noul": 0.0462,
+    "relational_compositional": 0.0012,
+    "score": 0.0785
+   }
+  },
+  {
+   "step": 4500,
+   "loss": 0.0635,
+   "tok_acc": 0.9759,
+   "entropy": 0.051,
+   "per_family": {
+    "choice": 0.0868,
+    "noul": 0.0527,
+    "relational_compositional": 0.0004,
+    "score": 0.0777
+   }
+  },
+  {
+   "step": 5000,
+   "loss": 0.062,
+   "tok_acc": 0.9757,
+   "entropy": 0.051,
+   "per_family": {
+    "choice": 0.085,
+    "noul": 0.0507,
+    "relational_compositional": 0.0022,
+    "score": 0.075
+   }
+  },
+  {
+   "step": 5500,
+   "loss": 0.0587,
+   "tok_acc": 0.9768,
+   "entropy": 0.0495,
+   "per_family": {
+    "choice": 0.0811,
+    "noul": 0.0516,
+    "relational_compositional": 0.004,
+    "score": 0.0672
+   }
+  },
+  {
+   "step": 6000,
+   "loss": 0.0553,
+   "tok_acc": 0.9784,
+   "entropy": 0.0571,
+   "per_family": {
+    "choice": 0.0778,
+    "noul": 0.0375,
+    "relational_compositional": 0.0013,
+    "score": 0.0685
+   }
+  },
+  {
+   "step": 6500,
+   "loss": 0.0554,
+   "tok_acc": 0.979,
+   "entropy": 0.0463,
+   "per_family": {
+    "choice": 0.0808,
+    "noul": 0.0349,
+    "relational_compositional": 0.0005,
+    "score": 0.0671
+   }
+  },
+  {
+   "step": 7000,
+   "loss": 0.0563,
+   "tok_acc": 0.977,
+   "entropy": 0.0443,
+   "per_family": {
+    "choice": 0.0804,
+    "noul": 0.0426,
+    "relational_compositional": 0.0083,
+    "score": 0.0621
+   }
+  },
+  {
+   "step": 7500,
+   "loss": 0.0506,
+   "tok_acc": 0.9797,
+   "entropy": 0.0564,
+   "per_family": {
+    "choice": 0.0698,
+    "noul": 0.0373,
+    "relational_compositional": 0.0006,
+    "score": 0.0633
+   }
+  },
+  {
+   "step": 8000,
+   "loss": 0.0511,
+   "tok_acc": 0.979,
+   "entropy": 0.0506,
+   "per_family": {
+    "choice": 0.0722,
+    "noul": 0.0344,
+    "relational_compositional": 0.0005,
+    "score": 0.0633
+   }
+  },
+  {
+   "step": 8500,
+   "loss": 0.0509,
+   "tok_acc": 0.9787,
+   "entropy": 0.0451,
+   "per_family": {
+    "choice": 0.0731,
+    "noul": 0.035,
+    "relational_compositional": 0.0006,
+    "score": 0.0613
+   }
+  },
+  {
+   "step": 9000,
+   "loss": 0.051,
+   "tok_acc": 0.9798,
+   "entropy": 0.0456,
+   "per_family": {
+    "choice": 0.0703,
+    "noul": 0.0399,
+    "relational_compositional": 0.0007,
+    "score": 0.0626
+   }
+  },
+  {
+   "step": 9500,
+   "loss": 0.0497,
+   "tok_acc": 0.9802,
+   "entropy": 0.0427,
+   "per_family": {
+    "choice": 0.0749,
+    "noul": 0.0319,
+    "relational_compositional": 0.0003,
+    "score": 0.0567
+   }
+  },
+  {
+   "step": 10000,
+   "loss": 0.0502,
+   "tok_acc": 0.9799,
+   "entropy": 0.038,
+   "per_family": {
+    "choice": 0.0699,
+    "noul": 0.0339,
+    "relational_compositional": 0.0004,
+    "score": 0.0637
+   }
+  },
+  {
+   "step": 10500,
+   "loss": 0.048,
+   "tok_acc": 0.9808,
+   "entropy": 0.0361,
+   "per_family": {
+    "choice": 0.0664,
+    "noul": 0.0295,
+    "relational_compositional": 0.0019,
+    "score": 0.062
+   }
+  },
+  {
+   "step": 11000,
+   "loss": 0.0471,
+   "tok_acc": 0.9815,
+   "entropy": 0.0378,
+   "per_family": {
+    "choice": 0.0621,
+    "noul": 0.034,
+    "relational_compositional": 0.0014,
+    "score": 0.0623
+   }
+  },
+  {
+   "step": 11500,
+   "loss": 0.0463,
+   "tok_acc": 0.9811,
+   "entropy": 0.037,
+   "per_family": {
+    "choice": 0.0644,
+    "noul": 0.037,
+    "relational_compositional": 0.0009,
+    "score": 0.0558
+   }
+  },
+  {
+   "step": 12000,
+   "loss": 0.045,
+   "tok_acc": 0.9821,
+   "entropy": 0.0387,
+   "per_family": {
+    "choice": 0.0621,
+    "noul": 0.0309,
+    "relational_compositional": 0.0007,
+    "score": 0.0572
+   }
+  },
+  {
+   "step": 12500,
+   "loss": 0.0466,
+   "tok_acc": 0.9815,
+   "entropy": 0.0341,
+   "per_family": {
+    "choice": 0.0667,
+    "noul": 0.0298,
+    "relational_compositional": 0.0007,
+    "score": 0.0574
+   }
+  },
+  {
+   "step": 13000,
+   "loss": 0.0444,
+   "tok_acc": 0.9829,
+   "entropy": 0.037,
+   "per_family": {
+    "choice": 0.0633,
+    "noul": 0.0292,
+    "relational_compositional": 0.0009,
+    "score": 0.0543
+   }
+  },
+  {
+   "step": 13500,
+   "loss": 0.0438,
+   "tok_acc": 0.982,
+   "entropy": 0.0383,
+   "per_family": {
+    "choice": 0.0607,
+    "noul": 0.0311,
+    "relational_compositional": 0.0011,
+    "score": 0.0545
+   }
+  },
+  {
+   "step": 14000,
+   "loss": 0.042,
+   "tok_acc": 0.9832,
+   "entropy": 0.0357,
+   "per_family": {
+    "choice": 0.0589,
+    "noul": 0.0298,
+    "relational_compositional": 0.0006,
+    "score": 0.0519
+   }
+  },
+  {
+   "step": 14500,
+   "loss": 0.0408,
+   "tok_acc": 0.9838,
+   "entropy": 0.0375,
+   "per_family": {
+    "choice": 0.0569,
+    "noul": 0.0309,
+    "relational_compositional": 0.0002,
+    "score": 0.05
+   }
+  },
+  {
+   "step": 15000,
+   "loss": 0.0398,
+   "tok_acc": 0.9852,
+   "entropy": 0.0374,
+   "per_family": {
+    "choice": 0.0551,
+    "noul": 0.0297,
+    "relational_compositional": 0.0001,
+    "score": 0.0495
+   }
+  },
+  {
+   "step": 15500,
+   "loss": 0.0401,
+   "tok_acc": 0.9848,
+   "entropy": 0.0327,
+   "per_family": {
+    "choice": 0.0542,
+    "noul": 0.0291,
+    "relational_compositional": 0.0003,
+    "score": 0.0518
+   }
+  },
+  {
+   "step": 16000,
+   "loss": 0.0385,
+   "tok_acc": 0.9855,
+   "entropy": 0.0356,
+   "per_family": {
+    "choice": 0.0529,
+    "noul": 0.0279,
+    "relational_compositional": 0.0006,
+    "score": 0.0487
+   }
+  },
+  {
+   "step": 16500,
+   "loss": 0.0383,
+   "tok_acc": 0.9853,
+   "entropy": 0.035,
+   "per_family": {
+    "choice": 0.0531,
+    "noul": 0.0271,
+    "relational_compositional": 0.0006,
+    "score": 0.0482
+   }
+  },
+  {
+   "step": 17000,
+   "loss": 0.0386,
+   "tok_acc": 0.9849,
+   "entropy": 0.0345,
+   "per_family": {
+    "choice": 0.0532,
+    "noul": 0.0279,
+    "relational_compositional": 0.0004,
+    "score": 0.0486
+   }
+  },
+  {
+   "step": 17500,
+   "loss": 0.0384,
+   "tok_acc": 0.9848,
+   "entropy": 0.035,
+   "per_family": {
+    "choice": 0.0533,
+    "noul": 0.0282,
+    "relational_compositional": 0.0004,
+    "score": 0.0477
+   }
+  },
+  {
+   "step": 18000,
+   "loss": 0.0384,
+   "tok_acc": 0.9848,
+   "entropy": 0.034,
+   "per_family": {
+    "choice": 0.0532,
+    "noul": 0.028,
+    "relational_compositional": 0.0004,
+    "score": 0.0481
+   }
+  },
+  {
+   "step": 18500,
+   "loss": 0.0385,
+   "tok_acc": 0.9853,
+   "entropy": 0.0342,
+   "per_family": {
+    "choice": 0.0533,
+    "noul": 0.0281,
+    "relational_compositional": 0.0004,
+    "score": 0.0481
+   }
+  },
+  {
+   "step": 19000,
+   "loss": 0.0385,
+   "tok_acc": 0.9853,
+   "entropy": 0.0341,
+   "per_family": {
+    "choice": 0.0533,
+    "noul": 0.028,
+    "relational_compositional": 0.0004,
+    "score": 0.0481
+   }
+  },
+  {
+   "step": 19058,
+   "loss": 0.0385,
+   "tok_acc": 0.9853,
+   "entropy": 0.0341,
+   "per_family": {
+    "choice": 0.0533,
+    "noul": 0.0281,
+    "relational_compositional": 0.0004,
+    "score": 0.048
+   }
+  }
+ ],
+ "best_checkpoint": {
+  "step": 16500,
+  "val_loss": 0.0383
+ }
+}