Add the completed calibration evaluation and the untrained control MMLU

.gitattributes

@@ -85,3 +85,5 @@ loopq_quantization/results/rtn_w4a4_group32/mmlu/harness/ByteDance__Ouro-1.4B/sa
 loopq_quantization/results/rtn_w4a4_per_token/hellaswag/harness/ByteDance__Ouro-1.4B/samples_hellaswag_2026-09-16T15-07-44.234921.jsonl filter=lfs diff=lfs merge=lfs -text
 loopq_quantization/results/rtn_w4a8_group32/hellaswag/harness/ByteDance__Ouro-1.4B/samples_hellaswag_2026-09-16T07-38-26.804149.jsonl filter=lfs diff=lfs merge=lfs -text
 loopq_quantization/results/rtn_w4a8_group32/mmlu/harness/ByteDance__Ouro-1.4B/samples_mmlu_professional_law_2026-09-16T11-50-42.373384.jsonl filter=lfs diff=lfs merge=lfs -text
+loopq_quantization/results/loopq_w4a4_completed/hellaswag/harness/ByteDance__Ouro-1.4B/samples_hellaswag_2026-09-17T00-06-20.554115.jsonl filter=lfs diff=lfs merge=lfs -text
+loopq_quantization/results/loopq_w4a4_untrained/mmlu/harness/ByteDance__Ouro-1.4B/samples_mmlu_professional_law_2026-09-16T23-55-25.767011.jsonl filter=lfs diff=lfs merge=lfs -text

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/evaluation.log

@@ -0,0 +1,217 @@
+INFO 09-17 00:08:36 [__init__.py:216] Automatically detected platform cuda.
+2026-09-17:00:08:46 INFO     [_cli.run:388] Selected Tasks: ['arc_challenge']
+2026-09-17:00:08:46 INFO     [evaluator:214] Setting random seed to 0 | Setting numpy seed to 0 | Setting torch manual seed to 0 | Setting fewshot manual seed to 0
+2026-09-17:00:08:46 INFO     [evaluator:239] Initializing vllm model, with arguments: {'pretrained': 'ByteDance/Ouro-1.4B', 'revision': 'e3b1e0993b1231a51d6069a870476dda4162c00a', 'trust_remote_code': True, 'dtype': 'bfloat16', 'tensor_parallel_size': 1, 'max_model_len': 4096, 'gpu_memory_utilization': 0.9, 'enforce_eager': True, 'max_num_seqs': 32, 'max_num_batched_tokens': 4096, 'enable_prefix_caching': False, 'seed': 0}
+2026-09-17:00:08:46 WARNING  [models.vllm_causallms:108] The `device` argument is ignored by the vLLM backend. To select which GPU(s) vLLM can use, set the `CUDA_VISIBLE_DEVICES` environment variable before running lm-eval.
+INFO 09-17 00:08:46 [utils.py:328] non-default args: {'trust_remote_code': True, 'dtype': 'bfloat16', 'seed': 0, 'max_model_len': 4096, 'enable_prefix_caching': False, 'max_num_batched_tokens': 4096, 'max_num_seqs': 32, 'disable_log_stats': True, 'revision': 'e3b1e0993b1231a51d6069a870476dda4162c00a', 'enforce_eager': True, 'model': 'ByteDance/Ouro-1.4B'}
+The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+INFO 09-17 00:08:55 [__init__.py:742] Resolved architecture: OuroForCausalLM
+`torch_dtype` is deprecated! Use `dtype` instead!
+INFO 09-17 00:08:55 [__init__.py:1815] Using max model len 4096
+INFO 09-17 00:08:56 [scheduler.py:222] Chunked prefill is enabled with max_num_batched_tokens=4096.
+INFO 09-17 00:08:56 [__init__.py:3400] Cudagraph is disabled under eager mode
+(EngineCore_DP0 pid=86391) INFO 09-17 00:08:56 [core.py:654] Waiting for init message from front-end.
+(EngineCore_DP0 pid=86391) INFO 09-17 00:08:56 [core.py:76] Initializing a V1 LLM engine (v0.10.2) with config: model='ByteDance/Ouro-1.4B', speculative_config=None, tokenizer='ByteDance/Ouro-1.4B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=e3b1e0993b1231a51d6069a870476dda4162c00a, tokenizer_revision=e3b1e0993b1231a51d6069a870476dda4162c00a, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=4096, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=True, kv_cache_dtype=auto, device_config=cuda, decoding_config=DecodingConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_backend=''), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None), seed=0, served_model_name=ByteDance/Ouro-1.4B, enable_prefix_caching=False, chunked_prefill_enabled=True, use_async_output_proc=True, pooler_config=None, compilation_config={"level":0,"debug_dump_path":"","cache_dir":"","backend":"","custom_ops":[],"splitting_ops":null,"use_inductor":true,"compile_sizes":[],"inductor_compile_config":{"enable_auto_functionalized_v2":false},"inductor_passes":{},"cudagraph_mode":0,"use_cudagraph":true,"cudagraph_num_of_warmups":0,"cudagraph_capture_sizes":[],"cudagraph_copy_inputs":false,"full_cuda_graph":false,"pass_config":{},"max_capture_size":0,"local_cache_dir":null}
+[W917 00:08:59.114198906 ProcessGroupNCCL.cpp:981] Warning: TORCH_NCCL_AVOID_RECORD_STREAMS is the default now, this environment variable is thus deprecated. (function operator())
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+(EngineCore_DP0 pid=86391) INFO 09-17 00:08:59 [parallel_state.py:1165] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=86391) WARNING 09-17 00:08:59 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer.
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:00 [gpu_model_runner.py:2338] Starting to load model ByteDance/Ouro-1.4B...
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:00 [gpu_model_runner.py:2370] Loading model from scratch...
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:01 [cuda.py:362] Using Flash Attention backend on V1 engine.
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:01 [weight_utils.py:348] Using model weights format ['*.safetensors']
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:02 [weight_utils.py:406] No model.safetensors.index.json found in remote.
+(EngineCore_DP0 pid=86391) 
+(EngineCore_DP0 pid=86391) 
+(EngineCore_DP0 pid=86391) 
+(EngineCore_DP0 pid=86391) 
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:12 [default_loader.py:268] Loading weights took 10.65 seconds
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:13 [gpu_model_runner.py:2392] Model loading took 3.0510 GiB and 12.431343 seconds
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:16 [gpu_worker.py:298] Available KV cache memory: 35.87 GiB
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:16 [kv_cache_utils.py:864] GPU KV cache size: 48,960 tokens
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:16 [kv_cache_utils.py:868] Maximum concurrency for 4,096 tokens per request: 11.95x
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:17 [gpu_worker.py:391] Free memory on device (44.12/44.42 GiB) on startup. Desired GPU memory utilization is (0.9, 39.98 GiB). Actual usage is 3.05 GiB for weight, 1.01 GiB for peak activation, 0.05 GiB for non-torch memory, and 0.0 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=38355684249` to fit into requested memory, or `--kv-cache-memory=42796003840` to fully utilize gpu memory. Current kv cache memory in use is 38512970649 bytes.
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:17 [core.py:218] init engine (profile, create kv cache, warmup model) took 3.87 seconds
+(EngineCore_DP0 pid=86391) INFO 09-17 00:09:17 [__init__.py:3400] Cudagraph is disabled under eager mode
+INFO 09-17 00:09:18 [llm.py:295] Supported_tasks: ['generate']
+INFO 09-17 00:09:18 [__init__.py:36] No IOProcessor plugins requested by the model
+2026-09-17:00:09:22 INFO     [evaluator_utils:446] Selected tasks:
+2026-09-17:00:09:22 INFO     [evaluator_utils:480] Task: arc_challenge (arc/arc_challenge.yaml)
+2026-09-17:00:09:22 WARNING  [evaluator:335] Overwriting default num_fewshot of arc_challenge from None to 0
+2026-09-17:00:09:22 INFO     [api.task:312] Building contexts for arc_challenge on rank 0...
+
  0%|          | 0/1172 [00:00<?, ?it/s]
  6%|▌         | 66/1172 [00:00<00:01, 657.18it/s]
 11%|█▏        | 132/1172 [00:00<00:01, 658.51it/s]
 17%|█▋        | 199/1172 [00:00<00:01, 660.75it/s]
 23%|██▎       | 266/1172 [00:00<00:01, 659.57it/s]
 28%|██▊       | 333/1172 [00:00<00:01, 661.80it/s]
 34%|███▍      | 400/1172 [00:00<00:01, 663.64it/s]
 40%|███▉      | 467/1172 [00:00<00:01, 659.43it/s]
 45%|████▌     | 533/1172 [00:00<00:00, 658.76it/s]
 51%|█████     | 600/1172 [00:00<00:00, 659.81it/s]
 57%|█████▋    | 666/1172 [00:01<00:00, 639.17it/s]
 62%|██████▏   | 731/1172 [00:01<00:00, 604.01it/s]
 68%|██████▊   | 797/1172 [00:01<00:00, 619.94it/s]
 73%|███████▎  | 860/1172 [00:01<00:00, 611.80it/s]
 79%|███████▉  | 925/1172 [00:01<00:00, 622.15it/s]
 85%|████████▍ | 992/1172 [00:01<00:00, 634.14it/s]
 90%|█████████ | 1057/1172 [00:01<00:00, 637.79it/s]
 96%|█████████▌| 1122/1172 [00:01<00:00, 640.84it/s]
+2026-09-17:00:09:24 INFO     [evaluator:589] Running loglikelihood requests
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+fatal: not a git repository (or any parent up to mount point /)
+Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
+2026-09-17:00:10:49 INFO     [loggers.evaluation_tracker:247] Saving results aggregated
+2026-09-17:00:10:49 INFO     [loggers.evaluation_tracker:119] Saving per-task samples to /workspace/loopq_data/curve_a4_s0/eval_final/arc_challenge/harness/ByteDance__Ouro-1.4B/*.jsonl
+vllm ({'pretrained': 'ByteDance/Ouro-1.4B', 'revision': 'e3b1e0993b1231a51d6069a870476dda4162c00a', 'dtype': 'bfloat16', 'tensor_parallel_size': 1, 'max_model_len': 4096, 'gpu_memory_utilization': 0.9, 'enforce_eager': True, 'max_num_seqs': 32, 'max_num_batched_tokens': 4096, 'enable_prefix_caching': False, 'seed': 0}), gen_kwargs: ({}), limit: None, num_fewshot: 0, batch_size: auto
+|    Tasks    |Version|Filter|n-shot| Metric |   |Value |   |Stderr|
+|-------------|------:|------|-----:|--------|---|-----:|---|-----:|
+|arc_challenge|      1|none  |     0|acc     |↑  |0.4770|±  |0.0146|
+|             |       |none  |     0|acc_norm|↑  |0.4906|±  |0.0146|
+

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/harness/ByteDance__Ouro-1.4B/results_2026-09-17T00-10-49.223641.json

@@ -0,0 +1,155 @@
+{
+  "results": {
+    "arc_challenge": {
+      "name": "arc_challenge",
+      "alias": "arc_challenge",
+      "sample_len": 1172,
+      "acc,none": 0.47696245733788395,
+      "acc_stderr,none": 0.014595873205358269,
+      "acc_norm,none": 0.4906143344709898,
+      "acc_norm_stderr,none": 0.014608816322065
+    }
+  },
+  "group_subtasks": {},
+  "configs": {
+    "arc_challenge": {
+      "task": "arc_challenge",
+      "dataset_path": "allenai/ai2_arc",
+      "dataset_name": "ARC-Challenge",
+      "training_split": "train",
+      "validation_split": "validation",
+      "test_split": "test",
+      "doc_to_text": "Question: {{question}}\nAnswer:",
+      "doc_to_target": "{{choices.label.index(answerKey)}}",
+      "unsafe_code": false,
+      "doc_to_choice": "{{choices.text}}",
+      "description": "",
+      "target_delimiter": " ",
+      "fewshot_delimiter": "\n\n",
+      "fewshot_config": {
+        "sampler": "default",
+        "split": null,
+        "process_docs": null,
+        "fewshot_indices": null,
+        "samples": null,
+        "doc_to_text": "Question: {{question}}\nAnswer:",
+        "doc_to_choice": "{{choices.text}}",
+        "doc_to_target": "{{choices.label.index(answerKey)}}",
+        "gen_prefix": null,
+        "fewshot_delimiter": "\n\n",
+        "target_delimiter": " "
+      },
+      "num_fewshot": 0,
+      "metric_list": [
+        {
+          "metric": "acc",
+          "aggregation": "mean",
+          "higher_is_better": true
+        },
+        {
+          "metric": "acc_norm",
+          "aggregation": "mean",
+          "higher_is_better": true
+        }
+      ],
+      "output_type": "multiple_choice",
+      "repeats": 1,
+      "should_decontaminate": true,
+      "doc_to_decontamination_query": "Question: {{question}}\nAnswer:",
+      "metadata": {
+        "version": 1.0,
+        "pretrained": "ByteDance/Ouro-1.4B",
+        "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+        "trust_remote_code": true,
+        "dtype": "bfloat16",
+        "tensor_parallel_size": 1,
+        "max_model_len": 4096,
+        "gpu_memory_utilization": 0.9,
+        "enforce_eager": true,
+        "max_num_seqs": 32,
+        "max_num_batched_tokens": 4096,
+        "enable_prefix_caching": false,
+        "seed": 0,
+        "config_source": "/root/venv-LT/lib/python3.11/site-packages/lm_eval/tasks/arc/arc_challenge.yaml"
+      }
+    }
+  },
+  "versions": {
+    "arc_challenge": 1.0
+  },
+  "n-shot": {
+    "arc_challenge": 0
+  },
+  "higher_is_better": {
+    "arc_challenge": {
+      "acc": true,
+      "acc_norm": true
+    }
+  },
+  "n-samples": {
+    "arc_challenge": {
+      "original": 1172,
+      "effective": 1172
+    }
+  },
+  "config": {
+    "model": "vllm",
+    "model_args": {
+      "pretrained": "ByteDance/Ouro-1.4B",
+      "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+      "trust_remote_code": true,
+      "dtype": "bfloat16",
+      "tensor_parallel_size": 1,
+      "max_model_len": 4096,
+      "gpu_memory_utilization": 0.9,
+      "enforce_eager": true,
+      "max_num_seqs": 32,
+      "max_num_batched_tokens": 4096,
+      "enable_prefix_caching": false,
+      "seed": 0
+    },
+    "batch_size": "auto",
+    "batch_sizes": [],
+    "device": "cuda:0",
+    "use_cache": null,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "gen_kwargs": {},
+    "random_seed": 0,
+    "numpy_seed": 0,
+    "torch_seed": 0,
+    "fewshot_seed": 0
+  },
+  "git_hash": null,
+  "date": 1789603726.0112321,
+  "pretty_env_info": "PyTorch version: 2.8.0+cu128\nIs debug build: False\nCUDA used to build PyTorch: 12.8\nROCM used to build PyTorch: N/A\n\nOS: Ubuntu 22.04.5 LTS (x86_64)\nGCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0\nClang version: Could not collect\nCMake version: Could not collect\nLibc version: glibc-2.35\n\nPython version: 3.11.14 | packaged by conda-forge | (main, Oct 13 2025, 14:09:32) [GCC 14.3.0] (64-bit runtime)\nPython platform: Linux-6.8.0-124-generic-x86_64-with-glibc2.35\nIs CUDA available: True\nCUDA runtime version: Could not collect\nCUDA_MODULE_LOADING set to: LAZY\nGPU models and configuration: \nGPU 0: NVIDIA A40\nGPU 1: NVIDIA A40\n\nNvidia driver version: 580.159.04\ncuDNN version: Could not collect\nHIP runtime version: N/A\nMIOpen runtime version: N/A\nIs XNNPACK available: True\n\nCPU:\nArchitecture:                            x86_64\nCPU op-mode(s):                          32-bit, 64-bit\nAddress sizes:                           52 bits physical, 57 bits virtual\nByte Order:                              Little Endian\nCPU(s):                                  96\nOn-line CPU(s) list:                     0-95\nVendor ID:                               GenuineIntel\nModel name:                              Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz\nCPU family:                              6\nModel:                                   106\nThread(s) per core:                      2\nCore(s) per socket:                      24\nSocket(s):                               2\nStepping:                                6\nCPU max MHz:                             3500.0000\nCPU min MHz:                             800.0000\nBogoMIPS:                                5600.00\nFlags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect wbnoinvd dtherm ida arat pln pts vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities\nVirtualization:                          VT-x\nL1d cache:                               2.3 MiB (48 instances)\nL1i cache:                               1.5 MiB (48 instances)\nL2 cache:                                60 MiB (48 instances)\nL3 cache:                                72 MiB (2 instances)\nNUMA node(s):                            2\nNUMA node0 CPU(s):                       0-23,48-71\nNUMA node1 CPU(s):                       24-47,72-95\nVulnerability Gather data sampling:      Mitigation; Microcode\nVulnerability Indirect target selection: Mitigation; Aligned branch/return thunks\nVulnerability Itlb multihit:             Not affected\nVulnerability L1tf:                      Not affected\nVulnerability Mds:                       Not affected\nVulnerability Meltdown:                  Not affected\nVulnerability Mmio stale data:           Mitigation; Clear CPU buffers; SMT vulnerable\nVulnerability Reg file data sampling:    Not affected\nVulnerability Retbleed:                  Not affected\nVulnerability Spec rstack overflow:      Not affected\nVulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl\nVulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization\nVulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop\nVulnerability Srbds:                     Not affected\nVulnerability Tsa:                       Not affected\nVulnerability Tsx async abort:           Not affected\nVulnerability Vmscape:                   Not affected\n\nVersions of relevant libraries:\n[pip3] numpy==2.2.6\n[pip3] nvidia-cublas-cu12==12.8.4.1\n[pip3] nvidia-cuda-cupti-cu12==12.8.90\n[pip3] nvidia-cuda-nvrtc-cu12==12.8.93\n[pip3] nvidia-cuda-runtime-cu12==12.8.90\n[pip3] nvidia-cudnn-cu12==9.10.2.21\n[pip3] nvidia-cufft-cu12==11.3.3.83\n[pip3] nvidia-curand-cu12==10.3.9.90\n[pip3] nvidia-cusolver-cu12==11.7.3.90\n[pip3] nvidia-cusparse-cu12==12.5.8.93\n[pip3] nvidia-cusparselt-cu12==0.7.1\n[pip3] nvidia-nccl-cu12==2.27.3\n[pip3] nvidia-nvjitlink-cu12==12.8.93\n[pip3] nvidia-nvtx-cu12==12.8.90\n[pip3] torch==2.8.0\n[pip3] torchaudio==2.8.0\n[pip3] torchvision==0.23.0\n[pip3] triton==3.4.0\n[conda] Could not collect",
+  "transformers_version": "4.56.2",
+  "lm_eval_version": "0.4.13.dev0",
+  "upper_git_hash": null,
+  "tokenizer_pad_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "tokenizer_eos_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "tokenizer_bos_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "eot_token_id": 0,
+  "max_length": 4096,
+  "task_hashes": {
+    "arc_challenge": "09f9ae87a0905d63512cffc4aa91a55e44258fc35160e40fa1eb66fb75473e34"
+  },
+  "model_source": "vllm",
+  "model_name": "ByteDance/Ouro-1.4B",
+  "model_name_sanitized": "ByteDance__Ouro-1.4B",
+  "system_instruction": null,
+  "system_instruction_sha": null,
+  "fewshot_as_multiturn": null,
+  "chat_template": null,
+  "chat_template_sha": null,
+  "total_evaluation_time_seconds": "130.1878583910875"
+}

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/installed_task_snapshot/README.md

@@ -0,0 +1,58 @@
+# ARC
+
+### Paper
+
+Title: Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
+
+Abstract: https://arxiv.org/abs/1803.05457
+
+The ARC dataset consists of 7,787 science exam questions drawn from a variety
+of sources, including science questions provided under license by a research
+partner affiliated with AI2. These are text-only, English language exam questions
+that span several grade levels as indicated in the files. Each question has a
+multiple choice structure (typically 4 answer options). The questions are sorted
+into a Challenge Set of 2,590 “hard” questions (those that both a retrieval and
+a co-occurrence method fail to answer correctly) and an Easy Set of 5,197 questions.
+
+Homepage: https://allenai.org/data/arc
+
+
+### Citation
+
+```
+@article{Clark2018ThinkYH,
+  title={Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge},
+  author={Peter Clark and Isaac Cowhey and Oren Etzioni and Tushar Khot and Ashish Sabharwal and Carissa Schoenick and Oyvind Tafjord},
+  journal={ArXiv},
+  year={2018},
+  volume={abs/1803.05457}
+}
+```
+
+### Groups, Tags, and Tasks
+
+#### Groups
+
+None.
+
+#### Tags
+
+* `ai2_arc`: Evaluates `arc_easy` and `arc_challenge`
+
+#### Tasks
+
+* `arc_easy`
+* `arc_challenge`
+
+### Checklist
+
+For adding novel benchmarks/datasets to the library:
+* [ ] Is the task an existing benchmark in the literature?
+  * [ ] Have you referenced the original paper that introduced the task?
+  * [ ] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
+
+
+If other tasks on this dataset are already supported:
+* [ ] Is the "Main" variant of this task clearly denoted?
+* [ ] Have you provided a short sentence in a README on what each new variant adds / evaluates?
+* [ ] Have you noted which, if any, published evaluation setups are matched by this variant?

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/installed_task_snapshot/arc_challenge.yaml

@@ -0,0 +1,3 @@
+include: arc_easy.yaml
+task: arc_challenge
+dataset_name: ARC-Challenge

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/installed_task_snapshot/arc_challenge_chat.yaml

@@ -0,0 +1,33 @@
+tag:
+  - llama
+task: arc_challenge_chat
+dataset_path: allenai/ai2_arc
+dataset_name: ARC-Challenge
+output_type: generate_until
+training_split: train
+validation_split: validation
+test_split: test
+fewshot_split: train
+doc_to_text: 'Given the following question and four candidate answers (A, B, C and D), choose the best answer.\nQuestion: {{question.strip()}}\nA. {{choices.text[0]}}\nB. {{choices.text[1]}}\nC. {{choices.text[2]}}{% if choices.text|length > 3 %}\nD. {{choices.text[3]}}{% endif %}\nYour response should end with "The best answer is [the_answer_letter]" where the [the_answer_letter] is one of A, B, C or D.'
+gen_prefix: 'The best answer is'
+fewshot_delimiter: "\n\n"
+doc_to_target: "{{ 'ABCD'[answerKey|int - 1] if answerKey|string in '1234' else answerKey }}"
+num_fewshot: 0
+generation_kwargs:
+  max_gen_toks: 100
+  until:
+    - "\n\n"
+    - "."
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+    ignore_case: true
+    ignore_punctuation: true
+filter_list:
+  - name: remove_whitespace
+    filter:
+      - function: remove_whitespace
+      - function: take_first
+metadata:
+  version: 1.0

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/installed_task_snapshot/arc_easy.yaml

@@ -0,0 +1,23 @@
+tag:
+  - ai2_arc
+task: arc_easy
+dataset_path: allenai/ai2_arc
+dataset_name: ARC-Easy
+output_type: multiple_choice
+training_split: train
+validation_split: validation
+test_split: test
+doc_to_text: "Question: {{question}}\nAnswer:"
+doc_to_target: "{{choices.label.index(answerKey)}}"
+doc_to_choice: "{{choices.text}}"
+should_decontaminate: true
+doc_to_decontamination_query: "Question: {{question}}\nAnswer:"
+metric_list:
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true
+  - metric: acc_norm
+    aggregation: mean
+    higher_is_better: true
+metadata:
+  version: 1.0

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/run_metadata.json

@@ -0,0 +1,724 @@
+{
+  "status": "completed",
+  "task": "arc_challenge",
+  "configuration": "loopq_w4a4",
+  "contract": {
+    "mode": "loopq",
+    "activation_bits": 4,
+    "artifact": "/workspace/loopq_data/curve_a4_s0/eval_final/artifact.pt",
+    "artifact_sha256": "7cb498143aeb807e5cb896f195066fa07e38a353406a10de36ffaf46b4fcabdb",
+    "ablation": null,
+    "parity_evidence": {
+      "path": "/workspace/loopq_data/curve_a4_s0/eval_final/parity_screen.json",
+      "sha256": "0492028e64eaf16242d69d32bec9a7783e3637875ddaae3b986889c86a1b04ed",
+      "acceptance": {
+        "passed": false,
+        "reasons": [
+          "four-loop hidden-state parity failed",
+          "logprob error exceeds tolerance",
+          "top1 or top-1000 comparison incomplete"
+        ],
+        "max_logprob_error": 1.25,
+        "max_hidden_error": 0.06,
+        "tie_threshold": 0.25,
+        "rows_excused_as_reference_tie": 0,
+        "scope": "six_or_more_single_sequence_prefills; decode/batch/long_context still require A40 validation",
+        "advisory": true
+      }
+    }
+  },
+  "model_args": {
+    "pretrained": "ByteDance/Ouro-1.4B",
+    "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+    "trust_remote_code": true,
+    "dtype": "bfloat16",
+    "tensor_parallel_size": 1,
+    "max_model_len": 4096,
+    "gpu_memory_utilization": 0.9,
+    "enforce_eager": true,
+    "max_num_seqs": 32,
+    "max_num_batched_tokens": 4096,
+    "enable_prefix_caching": false,
+    "seed": 0
+  },
+  "fewshot": 0,
+  "seed": 0,
+  "smoke_limit": null,
+  "study_evaluation": false,
+  "protocol_note": "LoopQ W4A4, calibration run to completion (1500 final optimization steps) with singular values held in log space so they cannot reach zero. The paper does not specify how the Kronecker transforms are kept invertible. Deployment prefill-parity gate is unreachable at W4A4 by any implementation and is attached as advisory. Explicit local protocol: zero-shot except MMLU 5-shot; plain likelihood scoring, no chat template; installed pinned harness prompts.",
+  "metric_note": "acc_norm where available, otherwise acc; retain all metrics",
+  "exact_paper_protocol_reproduction": false,
+  "disclosure": "Single explicit-seed run; paper reports ten-run accuracy averages. Dataset/harness choices must be compared before claiming reproduction.",
+  "command": [
+    "/root/venv-LT/bin/python",
+    "/workspace/LT_fix/scripts/run_lm_eval_timed.py",
+    "--model",
+    "vllm",
+    "--model_args",
+    "{\"pretrained\": \"ByteDance/Ouro-1.4B\", \"revision\": \"e3b1e0993b1231a51d6069a870476dda4162c00a\", \"trust_remote_code\": true, \"dtype\": \"bfloat16\", \"tensor_parallel_size\": 1, \"max_model_len\": 4096, \"gpu_memory_utilization\": 0.9, \"enforce_eager\": true, \"max_num_seqs\": 32, \"max_num_batched_tokens\": 4096, \"enable_prefix_caching\": false, \"seed\": 0}",
+    "--tasks",
+    "arc_challenge",
+    "--num_fewshot",
+    "0",
+    "--seed",
+    "0,0,0,0",
+    "--batch_size",
+    "auto",
+    "--output_path",
+    "/workspace/loopq_data/curve_a4_s0/eval_final/arc_challenge/harness",
+    "--log_samples"
+  ],
+  "environment": {
+    "torch": "2.8.0",
+    "vllm": "0.10.2",
+    "transformers": "4.56.2",
+    "lm_eval": "0.4.13.dev0",
+    "datasets": "5.0.1"
+  },
+  "returncode": 0,
+  "wall_clock_seconds": 139.81167713087052,
+  "source_provenance": {
+    "snapshot": "/workspace/loopq_data/curve_a4_s0/eval_final/arc_challenge/source_snapshot",
+    "sha256_at_launch": {
+      "loopQ/adapters/__init__.py": "7f28e14f7778bafb750404e0e54fb05fe17b61997e06584ca55ee44da3694ba7",
+      "loopQ/adapters/huginn.py": "fc22beafc461a7a0f85e6a7e3291a60749d5a11ea56cceabc448239dd3be9afe",
+      "loopQ/adapters/ouro.py": "e571df0cbbe853de8d1c593f9ed6c032e9099008363ee57e1f280b770ae49c14",
+      "loopQ/configs/flatquant_reference.yaml": "35f507d7389c3f04a7c1f86f7ae367630c846ec730661d420639f5f19b0c4074",
+      "loopQ/configs/huginn_mapping.yaml": "5db252e7cacb42169e94492e02509a3114ebca4747b32590110ba58cbbaea124",
+      "loopQ/configs/loopq_compatibility_decisions.yaml": "3d1b6d508db78b53f2c6e405da3b8ea6a7893303bd81bc2a90e0b3567367fd35",
+      "loopQ/configs/ouro_calibration.yaml": "fc832996fcdda085290565b657c6fa1824dc0b1646a66d109fe38b22f0a8eac5",
+      "loopQ/configs/ouro_mapping.yaml": "87ca4718e200539412f366fbad52a1a46932ad2cc26e0f08585830d04f50b9ad",
+      "loopQ/configs/paper_ouro_reference_metrics.json": "ecc4b3e028b1191a458b446582895d4f6871922a397f15375a34e396bf70ad89",
+      "loopQ/configs/paper_specification.yaml": "c812bb40398d9183f522c07e32dabf3fcfdd43e4d52790b9af296fb11d5abed1",
+      "loopQ/configs/parity_prompts.json": "b56e2e4048083459997c2864e3e6320fc68e1625676ed09f207a81bafee9b2e0",
+      "loopQ/configs/reproduction.json": "5204e346d98f96e46b0f94814e2f291198887ace85a7c827932f002560c350c3",
+      "loopQ/loopq/__init__.py": "00cf79e9673051148513cfc25c52c59ef72118d288e2ec640a77eaf8a69d7213",
+      "loopQ/loopq/accounting.py": "ab24ef8b10a0bc85e03827e882f92e1d7bb3e33fbd146f7f9561651842a03fb0",
+      "loopQ/loopq/calibrate.py": "5018098ddf0c44a682efef85b9f82ab2e4f4f911efa48982083d6c3a0db98594",
+      "loopQ/loopq/cta.py": "e1cf8cfec967157a90d9f54e956b221b67aef9ed742968605d2d4abb1f453b8d",
+      "loopQ/loopq/evaluation_provenance.py": "87e0d6f329dae65759893063cf508682902ae4cf5772aec28ed1692a08da02cf",
+      "loopQ/loopq/experiment_contract.py": "a0f4487092a72eb5fe808880d09a6fafc9b86e5fc4a2e2b3b9e7d5177d27a620",
+      "loopQ/loopq/experiments.py": "252644613f076890e5f2b849c4f95570e33a9cfa69569df90b082c26788f5a5b",
+      "loopQ/loopq/export.py": "3ca272d220bbc93d2e7c70b06df2cef93184e48bc695c00b97e834991469ef57",
+      "loopQ/loopq/gpu_admission.py": "feb940b5cc2355f1881d031f9b22226d7692f70f792c1b1233aeec5570535c76",
+      "loopQ/loopq/huginn_calibration.py": "7a95635b9d9777c9b0f04f51566f9ed668154ae9be231e19f3b7ac18acef98ca",
+      "loopQ/loopq/journal.py": "f9bf8342a004959bbdb25b3557577e5271e7335f53107d7a007ad512fa55bbcf",
+      "loopQ/loopq/las.py": "a179c62e57fccafc21dabb13a68832f8bc8c9d6b0a8da1a2c2348bfb09c47a3c",
+      "loopQ/loopq/numerical_contract.py": "913685841222a9f7ff503ca9b8962cd976f39bda38e2a20f7d3e6b0c09798bc5",
+      "loopQ/loopq/objective.py": "d60f1d6ce18e004f89cb91f680ed6dd4b28083dda8125745a1bd694389c40e50",
+      "loopQ/loopq/ouro_calibration.py": "db88d48b9646c421e4fd354e8de61a7784bdedcbac0b47e1c8de01aa8c2395de",
+      "loopQ/loopq/packed_bundle.py": "5fe29dee51fe28849696de298d3bc8f7036b78c44ec2617747ada370e135287f",
+      "loopQ/loopq/packed_weight.py": "8bbd5a3dc0c8992e4dd30ce962deb0e0745604b1e189a508e1f70aee1e1acdbc",
+      "loopQ/loopq/packed_worker.py": "239febb2ed17d9d88cb068d2edac87f13c2f5c579995a842cdfb7ab6b1e536aa",
+      "loopQ/loopq/parity_worker.py": "b6854262f3b44e4970214aa3e8ccc5a8ad1fa032a9ae5314a15514e132540502",
+      "loopQ/loopq/paths.py": "5a9478359f3a4b252d5af599c9c7f68650d10b995f2e9314210ac6e670a364d0",
+      "loopQ/loopq/provenance.py": "387f82f3efe26ea478476d7087db441a2552f345de35ce2afc5723eb4d21abd6",
+      "loopQ/loopq/quantization.py": "78a3016ad8b2d53ae807bb18b71da8e7d8ce6df8d0881331a80776a247c1142e",
+      "loopQ/loopq/readiness.py": "837a59bc6ac641ea2140876fd3ae34afc30173b91791852ba55dafd4bdd9cdd8",
+      "loopQ/loopq/saved_tensors.py": "ee158cc09d8116c09922b940d3219d7d22d3f271b9c11e8b623b393cecb2af03",
+      "loopQ/loopq/sharing_gap.py": "aa6e1bdb0f365fdde0e9ed4674a08b0f2c0df0a0693dd28dc5c094f8f86118a5",
+      "loopQ/loopq/trajectory_diagnostics.py": "de78ef945d1fdcd036460ad5947864f73a5d63a5f88a1a6c86691821b2dddcac",
+      "loopQ/loopq/transforms.py": "d754f21df55fde2548872d25afa9a4fe75082fdc937a3ba65b474a0673be12a6",
+      "loopQ/scripts/benchmark_calibration_memory.py": "afdaf39cde1e839924baf1cfd0a4d83df5ee0af40d6e7659cb6f3645e3ebb408",
+      "loopQ/scripts/calibrate_huginn.py": "ba44cc894b5a9987cb2f3a1382053d907f205a2d7608b44b7714fc410c142756",
+      "loopQ/scripts/calibrate_ouro.py": "5c81cb8ca6a92a6da4fe1b7e5d7615830e359d93a4eedf013fbb32d5f624ec9e",
+      "loopQ/scripts/diagnose_ouro_quality.py": "962f5ac918bea5cb029c78f41b96b886013e8233d5bf1d67ec87d8b488b9977c",
+      "loopQ/scripts/export_huginn_artifact.py": "20dbf608320f85daddbc843b0ac1196234c20822e2418f42b66327ddcd8186f7",
+      "loopQ/scripts/export_ouro_artifact.py": "1fe2edfa0fc09cf896a1fab149067160829744d3f7fd03d2c521b6ba38d19a12",
+      "loopQ/scripts/export_packed_ouro_weights.py": "a49b8646639682012340f908ec86efbe12a369343c984963c2314991c31a4ead",
+      "loopQ/scripts/freeze_lambada_inputs.py": "b15138ae82a52461f31cbb669d18f878fd98f9eab93ac7db3efa87f594afb97b",
+      "loopQ/scripts/inspect_calibration.py": "6514bb2fb9fda6ef6b5335035dae910cb4bf7ec36dde5a844c7cf38e0fd07f7a",
+      "loopQ/scripts/measure_ouro_trajectory.py": "06bf7ea56bb6626b5947e8f43d2b7ddc60ad942d3997ac1c2b7795db4672c081",
+      "loopQ/scripts/report_accounting.py": "3564bf5cb9e4f8133257cb415ec679f007090f0c7afe85fe1bca4b469b5ba5e6",
+      "loopQ/scripts/reproduction.py": "6a28f7a4cb1f4a9f4171fcf466b36b103761bc38bde63e3afefe8a25aca86fb1",
+      "loopQ/scripts/run_gsm8k.py": "11b220ad74277b5da6becbf18eaaeb4725d6ecb07330b1b15cd4dc4c29eeefdb",
+      "loopQ/scripts/run_paper_benchmark.py": "021ee1451abff633bc0b77f04c13e1cd9dae7bebfc86ebe962890015a74a6bf8",
+      "loopQ/scripts/smoke_huginn_vllm.py": "9311b038ce325a3f050a852addb6a9c550a09ace19f29ee3370439be9047e29b",
+      "loopQ/scripts/smoke_ouro_vllm.py": "39311a193900e7d1db491e29fcdd83ee88ad1bdda602025580f61bd5ae757203",
+      "loopQ/scripts/verify_runtime_parity.py": "1e0e1648fe5673abaca26d98efffac99cc3b2ede7ecb59423df04147e537e4e0",
+      "loopQ/tests/conftest.py": "56cfc62ddfe4ccab80288f60c8f3948b6d8b8d8f1d2e37c50f9c9577c4679235",
+      "loopQ/tests/test_alignment_contracts.py": "615d2e77c187411d55892fa875835ea280a9b6ab775835ed30459482c298f066",
+      "loopQ/tests/test_calibrate.py": "d9bddf96ebcd0c5bdd0dd866f17e045d672abd73360e5e0771775edc4e67c8c8",
+      "loopQ/tests/test_cta.py": "afe35581d60dcd0b5ff1566ffe3a8dc332003cd031d9c9e788cff1c2f62f2307",
+      "loopQ/tests/test_evaluation_provenance.py": "62765331e2e5b0e400c51b0d7caba5cc70f3060ac60fdfa53e89fc0eea293933",
+      "loopQ/tests/test_experiment_setup.py": "840680d43db300613ccb55409ce9bd142f2eb82176d697813194648de2c57484",
+      "loopQ/tests/test_export.py": "7bf542a1e606c7be517c2485dcad3b8e8ec773f53e7374656564ce33a78a6b8a",
+      "loopQ/tests/test_gpu_admission.py": "5f0154e0e7d80e1de684b51219b3ee1b52d81481d062de13a563eab1db17fce6",
+      "loopQ/tests/test_gsm8k_driver.py": "054d0284905dc8479ceb6c28e2a418d94a692762d45573902a8bb0278abd2821",
+      "loopQ/tests/test_huginn_adapter.py": "23f7b5d2fea8c9526553e979a5c44e466fe74d664cf6f0a05b8c66434dc48940",
+      "loopQ/tests/test_huginn_calibration.py": "bbdc7f0e10f23042f6d52acf14456d039b030d945fb4d3a8c72d9d6063242a93",
+      "loopQ/tests/test_huginn_export.py": "896eee3b771ee2a2b0dbdbd9aa07eb9918718c7dc9e69e0169b3d4ab22e50229",
+      "loopQ/tests/test_huginn_vllm_bridge.py": "c483f52999ca955b4c8b17de252e957bf7a878bfa7e5fa18330c7fed3f249c9b",
+      "loopQ/tests/test_journal.py": "b3331cbff1add620c8e1d098578a9d79f81e2c1d28a2e0c0b4a7a7a0c7ba1589",
+      "loopQ/tests/test_las.py": "88139ae1f23a53d07eb3497b88fe3f22980d6b8c001232fc3b629a57663935e4",
+      "loopQ/tests/test_objective.py": "fb206b4844e1f911e6665ccd5d99e643171810a30da50a80be64d227019e9b88",
+      "loopQ/tests/test_ouro_adapter.py": "9a52daa93ea65a75fc9edfd154ab27058c6a2567d113cab25e6d5a9e6592d986",
+      "loopQ/tests/test_ouro_calibration.py": "e9099d1cddec60482c811df656e7aa8731c2a1cd98ae274efde249636d0cec50",
+      "loopQ/tests/test_packed_bundle.py": "cfffba2fd0d640e2e5def1d6375b24a87b4d33bb57a7c45fb5d08145ae46bd4d",
+      "loopQ/tests/test_packed_weight.py": "387fdb51e3bce2cbd7c0f94dcec5fb1ecd8eedd55c79b6811975503625be884f",
+      "loopQ/tests/test_packed_worker.py": "41206a263d72bdc2e7ba65c8ecf7e6982f3b31b62c9602e554798212f8373a37",
+      "loopQ/tests/test_paper_benchmark.py": "70eaa2326bf0b8da12736a1577421a42af446aacd48fc21be0e9ead2deab52ed",
+      "loopQ/tests/test_paper_math_reference.py": "a46176f310f90d75bef72674620a2425de13a20bb65f84d33686dc82ff4043fb",
+      "loopQ/tests/test_quantization.py": "a0368909de90a7b5d64d0cb82296ab438306f4e337c50ed1a103616fdd3f9af4",
+      "loopQ/tests/test_saved_tensors.py": "522eae339ca2d2b01506f653b18d6417bc04b4315d0d1afa90b78820d55c7742",
+      "loopQ/tests/test_sharing_gap.py": "aacbacadaee3850d1d823cf8122cbc2ac80f2af3066254a7b5b60d46c0ded1ac",
+      "loopQ/tests/test_trajectory_diagnostics.py": "341d8650a0c3ae59c12584baaccc6fec454bb76f988e533385f5953d3f83bf3e",
+      "loopQ/tests/test_transforms.py": "5e9fc5b387fb8f40f3b090864a0c3751529a869bef176058bac149dd4401bfca",
+      "loopQ/tests/test_vllm_bridge.py": "ccab13d1a9c3bd18279e46b4268cccba84e766153282e9d78cd58e093fe50647",
+      "src/loop_transformer_quant/__init__.py": "a9dceaa469212264a6ee7b1a615de2c5cd6e6195a007c852fc7e1e85857c172f",
+      "src/loop_transformer_quant/adapters/__init__.py": "6d112e61f7a6dc6093f11278d03a5351e8dd7adcb55d34dbb9e6810f86dbee74",
+      "src/loop_transformer_quant/adapters/vllm.py": "9f8df73df24a2b2297def5f0b994b8508fd06dc1a16184af64220e9016b813fd",
+      "src/loop_transformer_quant/cache_codec.py": "cf22a2cda1546e2bacb1e7e6ce9d8bdf8db952d758dd5cf81fdc1e784154185e",
+      "src/loop_transformer_quant/dynamic_lse.py": "ca0ce130dd1f838ab34c96ba8fe23f308ba09975801f587e4d5fa5723ddd93f5",
+      "src/loop_transformer_quant/gsm8k_calibration.py": "f90b5b89e595ef86ca6d03ae4f77bfa62b123aee3b9efcacfa468a87c72d64fc",
+      "src/loop_transformer_quant/huginn_anchor_mechanisms.py": "bb59f9f3d90ff0bf3dada6cd7d48ddf26169796a0553a4c2a39fd520c2222b8a",
+      "src/loop_transformer_quant/int2_experiments.py": "7be8a00a688a57ee7aa0e8fbfe6e8adf15277dcacabd95e608b5707d9010abba",
+      "src/loop_transformer_quant/int2_runtime.py": "2eb7f992fe80c63874efb440f8ed3b1c84374d16118872532d9ada3cde3a0b1e",
+      "src/loop_transformer_quant/int2_streaming.py": "e16f5ab4abd71af8e3ea5ec3cc3164a5956bacbe87cd81f1b7c2af551bb2bd75",
+      "src/loop_transformer_quant/kv_codecs.py": "118c98e0e32d899d7e6458d8c626738912af9ce87d761789a0d7d2a51e8d24c6",
+      "src/loop_transformer_quant/layer_channel_heatmaps.py": "04f09e794e6f87d75410a2ff55bbd60f047bdaf5355dcb4abca085b0a0fb318f",
+      "src/loop_transformer_quant/logical_storage.py": "01858edbf3c5406af60c747375c0fbee6b406fc867a0ea1741525caeb90245f3",
+      "src/loop_transformer_quant/native_activation.py": "4f82f90e80574e92e2dd97128a206983f32e0bb424286fad33ceee91c19dd366",
+      "src/loop_transformer_quant/native_kv/__init__.py": "acfa90ad9c90018df7f68d2efecfd8f9dbb4ca90344279ba3aa307e6829a5b39",
+      "src/loop_transformer_quant/native_kv/cache.py": "573b46adf72ad17415ad2e1a8af977cefa63f3958698fee7ecdc50b2a8a8ff2a",
+      "src/loop_transformer_quant/native_kv/calibration.py": "6b5123c4933e147f44c8a47c339311a034f842957155894f81eeea0f4be0e674",
+      "src/loop_transformer_quant/native_kv/channel_lse_certified.py": "43d0d6dcce1de9fff4c99979336a0861c67f7bf6d7ba786509201ca03e8cd0cf",
+      "src/loop_transformer_quant/native_kv/channel_lse_cost_probe.py": "8a816598f39847d32318383e456dab2e575e04047ed89434fcc830428ece23f8",
+      "src/loop_transformer_quant/native_kv/channel_lse_fastproducts.py": "c7b4b4927725c7388342aee92abf014304e81714c27050bce4709725a0a3eb19",
+      "src/loop_transformer_quant/native_kv/channel_lse_guarded.py": "04f2893e42049ed01b7c0c4a4076f9ae5ad786562fc81311e2ab6ffbdca0fefc",
+      "src/loop_transformer_quant/native_kv/channel_lse_kahan.py": "036991681f5e6667f6684c0b794eefdfdcbdffa2b096095413544a848c96ba00",
+      "src/loop_transformer_quant/native_kv/channel_lse_mixedstats.py": "3acb22432e21870c99d83f12121cec2882bff1ad9b14ebc27600699b68ea9cc5",
+      "src/loop_transformer_quant/native_kv/channel_lse_native.py": "fe7d3082aba93d483983e091053d722860b51b478e138f92a27114c5c6ee7a11",
+      "src/loop_transformer_quant/native_kv/channel_lse_pagemixed.py": "2b83812cf5133e4a88780d4a89fa7179ec1ea581f68b4f356551d896e162a981",
+      "src/loop_transformer_quant/native_kv/channel_lse_pagetiled.py": "5f38be7492b93176cbe998d3b33345056a6db90c40275261b5c65ce3631cf104",
+      "src/loop_transformer_quant/native_kv/channel_lse_reference.py": "b09e33d551bbe31a6fd56469a207f8a6c9173eb915761c3c263c2574ee31567e",
+      "src/loop_transformer_quant/native_kv/channel_lse_solve32.py": "eeafd3685da864208cffefef9d5bd3575a8a2b873baf7f38e79dcbcf3fbd64f9",
+      "src/loop_transformer_quant/native_kv/channel_lse_split_solve.py": "d847cccf05382f38c337c094654e892c1544e37dff1fbb89967e349220756ccf",
+      "src/loop_transformer_quant/native_kv/channel_packed.py": "083f30a6446e125079df2e17811d06d004ee289953b999304f95e989acd86513",
+      "src/loop_transformer_quant/native_kv/channel_paged_fused.py": "a7fc2ab879859804913838b38c7888db589420f2fcdf5303165ff6a72c4d6f57",
+      "src/loop_transformer_quant/native_kv/channel_paged_grouped.py": "e52cf976ee34bd680af8107b18e97e50b7272deac2b3925e5c12f1d248171ca7",
+      "src/loop_transformer_quant/native_kv/channel_paged_meta.py": "179c3df4189e034f5c5b233f23e76bc0a49d12bab990480525d5f033ecc2ea6e",
+      "src/loop_transformer_quant/native_kv/channel_paged_recompute.py": "0ca703afd01d66337c949af2b3d73513bf801cfecb41f1c7851120e7e6641f63",
+      "src/loop_transformer_quant/native_kv/channel_prefix_scan.py": "3ef7845af24a20f515b2f04461c97fb2eed3d9d9200cc60d77bca4b7b34b0c1b",
+      "src/loop_transformer_quant/native_kv/channel_split_write.py": "8b197f54626758564c220aba8908de298ae2f4929a7131abbdf2d9fdea192421",
+      "src/loop_transformer_quant/native_kv/channel_token_fused.py": "eb455108f066f1ad90821a9484efc2ac4967f1b72d5ac83773054ff36ab16f36",
+      "src/loop_transformer_quant/native_kv/channel_token_meta.py": "443f3332b6e31cba118e880f71fed429c007f6d53543625d273b66eef4fe90a8",
+      "src/loop_transformer_quant/native_kv/comparison.py": "d318411d27db8f903ca1f250db8196d3b490e7892f9ee49d647dc3f304f88b00",
+      "src/loop_transformer_quant/native_kv/dynamic_paged.py": "c34a21557bd218278b6a08a8cdaa0b0a30a2b29a1ff02db02579f9cb7c73556b",
+      "src/loop_transformer_quant/native_kv/dynamic_paged_fused.py": "3c0f5c620c6bcb938bea32a94aa832c70fb356716088dfda207f47004727e393",
+      "src/loop_transformer_quant/native_kv/dynamic_scalar.py": "69e954ac3100cddb7464ec3e6caf9c3d3927fc96b99c472868057aa821a69867",
+      "src/loop_transformer_quant/native_kv/fa_tile.py": "5836fc96d54b5892fa911f0d0d48f81c9a91d71690c0455be639f31958bd77d0",
+      "src/loop_transformer_quant/native_kv/kernels.py": "d5537c45945724076984af5ff68c9e7b859528bf5c6ccc437c3885fc96f002f0",
+      "src/loop_transformer_quant/native_kv/ouro_comparison.py": "bf232a254102672863fe233e627a4e0059286a7f55e6362e397b22ce83178e77",
+      "src/loop_transformer_quant/native_kv/request_slots.py": "662be79c9ca67d8dace1c843a4ccb1b9325cb76b0b97c9c1fcdc677e44e86e25",
+      "src/loop_transformer_quant/native_kv/static_paged_fused.py": "19dfc2d58beac8b5318a97ca8fa2b5ac3badc4f33c0ea013d70e9db4d33d8466",
+      "src/loop_transformer_quant/native_kv/xqa_loader.py": "4de9c6c9f3a68d4fc422d68d3e55868dc756686299f901127acf284148472b0e",
+      "src/loop_transformer_quant/nvfp4.py": "68f0623c31f9667b295d09b2ecf4832aee84dbf93ab3de7348579189ca44a7d1",
+      "src/loop_transformer_quant/packed_affine_kv.py": "3c249789ca964aa1947262e9177fc3f66401d67889d1396a511cc489ea89e682",
+      "src/loop_transformer_quant/packed_rotated_hnd.py": "844421a98211c9a4bc1867c79f19c7d5583e82abe606b941276260c9a9e9c662",
+      "src/loop_transformer_quant/packed_rotated_hnd_safe.py": "9409fd0cca643ed8b3c87de99440e66eb7d3e9e8d6fdce8c76553806c3c29323",
+      "src/loop_transformer_quant/packed_rotated_kv.py": "195427fea9b110b693fbb54a1cd809482bdec7c75417055ec9ad14c8ed182ecb",
+      "src/loop_transformer_quant/packed_rotated_table1.py": "9b246daf90bf8d005482d413fd4f72422945d8ac001e98c61075365a6e1e8885",
+      "src/loop_transformer_quant/paths.py": "1f87fce7344da5916fd4a8b88dec2f5d7b45bf4b4cab129b72e9bf2280595d9c",
+      "src/loop_transformer_quant/pre_sinq.py": "e7926059546171473b26d1fdcfb2378884fd95aefe573d93119acf981e13240e",
+      "src/loop_transformer_quant/quant_policy.py": "bc55de3e7d27d7825cb2dbf443cd90ab3fe0452856a38864ecdb93c2e0fe3e38",
+      "src/loop_transformer_quant/rotated_attention_adaptive.py": "6ca61bf1207abe12e40de81900348df49d1832a69d1d2b9b92fee81daceedc34",
+      "src/loop_transformer_quant/rotated_attention_adaptive_safe.py": "5c93b4f091f723a1631b210bfb0589501c191d87adc57eacc8e90c98ae3b3f2c",
+      "src/loop_transformer_quant/rotated_attention_combined.py": "24d0fea4c099321594fe0d80c82093e934fd231137559d1a9d90535470db0458",
+      "src/loop_transformer_quant/rotated_attention_combined2.py": "207860a760e4a2bb2491d233247fabe965c84796ad7eb1bdab83df9a64d3d77c",
+      "src/loop_transformer_quant/rotated_attention_combined4.py": "df49a846ac8059bbdd9e5441fa91d366005063d0946c1c1fafff929998b5ca06",
+      "src/loop_transformer_quant/rotated_attention_conservative.py": "6318dbf853294db80b397985b0f3b92fce7d97791c752a0fb9df949095b4a294",
+      "src/loop_transformer_quant/rotated_attention_fast.py": "2e884106998c22e14fe822d3267478da54521ebfb93ab76e174daec23913ba9f",
+      "src/loop_transformer_quant/rotated_attention_group_all.py": "d6973ecff1f66eed99f9955b102f6ba5d51d240e249f724d6667028998da0347",
+      "src/loop_transformer_quant/rotated_attention_group_codes.py": "be0723b591b7c95436057c3ea7616c4178e516c8dc80495b698375196239e921",
+      "src/loop_transformer_quant/rotated_attention_group_codes_v2.py": "327119e221287a8221d178290aba72c4d428ff1d6664347374f04d287146da83",
+      "src/loop_transformer_quant/rotated_attention_group_load.py": "7cdda56e000afe900d7b6d3a942f112c4ab986a1a70476275531dc3238696063",
+      "src/loop_transformer_quant/rotated_attention_group_pipeline.py": "b2139be31958bb37fdf0afbfdfdbcde8ab91245208782ba16631ea47b369f3f3",
+      "src/loop_transformer_quant/rotated_attention_group_rotunroll.py": "f4bdc70394069dc078649b6ad5310ccbce2e5d9dff048f403fcebba687ce6fb5",
+      "src/loop_transformer_quant/rotated_attention_hadamard.py": "2dc3ea5984f251fb17d93c4868cdc6297801723bc6799ac1a9d4d96bc18e46e2",
+      "src/loop_transformer_quant/rotated_attention_pipeline2.py": "81efddc54396b60dd7c2c412e092e43eddacb0d9641f2d8a4eec9878d72eb4ee",
+      "src/loop_transformer_quant/rotated_attention_query16.py": "161faf38eba65c083fac344853dfade1ec326da7d6257469d10626c2abce5df1",
+      "src/loop_transformer_quant/rotated_attention_query8.py": "143350254d30170998b5300874e1c04a44adc5e3cf23563b33fe61dd1e8c77c7",
+      "src/loop_transformer_quant/rotated_attention_rot_keep.py": "2f55a8ed07a478067ae68b911ee457b4f0628beeae078b121eaae0c8cf5d14e8",
+      "src/loop_transformer_quant/rotated_attention_rot_transpose.py": "dd7823da1e66c88a4f7b74874d25d94261cbc17e85fb02a73938fcfce6fb7765",
+      "src/loop_transformer_quant/rotated_attention_unroll2.py": "90bb8c89664acff921d0369083aa7f53ed628b0168ff5d2e8779654fe727ea97",
+      "src/loop_transformer_quant/rotated_attention_unroll4.py": "4cd2474e5a63dd893d3fc545d12fdf0ea2d2be0e4d232bed1130b5aa7ce95539",
+      "src/loop_transformer_quant/rotated_attention_warps16.py": "c984459e4fc0d9d1d3d0bfe8ec41f1abb7e9ad723da30506ba44a68ca25085f7",
+      "src/loop_transformer_quant/rotated_attention_warps4.py": "44401f179631c6e6120b2e111326e77c14a2edd002ec28d3f8168d102d22695e",
+      "src/loop_transformer_quant/rotated_attention_xqa.py": "e7a8a71abac60a1865a092d25777a499b4fbb6ba8722430c0d4e705f28f4f786",
+      "src/loop_transformer_quant/rotation_policy.py": "7fb06183b276662564c2f64f14bd7ea6a5438b46b3d8a05efc3678f46c4b54cb",
+      "src/loop_transformer_quant/rotation_study.py": "c05e433e8182e7af43ae499d09fc5eb5f8a53512cef4d01be9dd2d59ea26e14e",
+      "src/loop_transformer_quant/rotation_table1.py": "c7bbd247eab272600bcba2219c83160bb54abc4bcff3d1de8ca496fe2faff985",
+      "src/loop_transformer_quant/runtime/__init__.py": "c412e782654be12e77250c6b617b5034c9a2e33a1084b7880f8c0eacb46e05f8",
+      "src/loop_transformer_quant/runtime/cache.py": "2e1f0026738ab93ee3767ca29a53ff38a67d707f3e64f466fc402a80eebc82fa",
+      "src/loop_transformer_quant/runtime/context.py": "e8f723bc270d1293b086d621c66f69a964e570175df733e23c9755e74933ca10",
+      "src/loop_transformer_quant/runtime/module_cache.py": "f189187bbbc1436049ba863550ac0b4134466e09bc1ee3471c3b030b33d01e17",
+      "src/loop_transformer_quant/runtime/specs.py": "b9a3deee69e44c8ccc009e92bdf34c83662b7d553ffebe25eabbead0b897bdfc",
+      "src/loop_transformer_quant/sensitivity.py": "00d0ab71278e19113dd6b5db65d091271fe1be65054562677edf7c082d8f3022",
+      "src/loop_transformer_quant/sm120_fp4.py": "301b7a8141337b68fda40dd9fdeba2d4eac46fbb17c0d61292481910ece83ca4",
+      "scripts/account_recurrent_kv_storage.py": "00aca9766fa7960c010b5170f36781224e45a4a288518bf37c6864b14422fbde",
+      "scripts/analyze_accuracy_grid.py": "0d33123eae6c5beaf625854f6d0b92e4169a835e8fb067139c0b037dcfcf2e46",
+      "scripts/analyze_huginn_anchor_ablation.py": "3d825b07b42a08a8f9dbb8fe7b39bb50c8b51c5c7f02da8183ad11a90d43906a",
+      "scripts/analyze_huginn_full_validation.py": "6b8a5972ff7f878d2368273a41898ccbd5321e61056a59c8ac845fd2f01a7741",
+      "scripts/analyze_huginn_mixed_precision.py": "d8cd6ee6c8f39ec28fffb6ebbdc618443bc88918dd1c5ebe06a8ebac30289519",
+      "scripts/analyze_huginn_residual_kv.py": "015e3bd1abbdc14c98b077d645609ef4fc312654a8c854f421a62e0516d467df",
+      "scripts/analyze_int8_scale_underflow.py": "181ea6a32f88a9edaab237edaa5fcb02d5baa85d518ce996a53ffba214601388",
+      "scripts/analyze_int_scale_matrix.py": "b0d17fc01e36e0cae0bb2fcd12b67e578b0b67983159ee5dbf582cd01a5c7b12",
+      "scripts/analyze_kv_codec_screening.py": "c712a196068fa991a2272435d5c302edd570fd4100209dcd5bf316103557ba35",
+      "scripts/analyze_loop_delta_quantization.py": "dd1201e93c2183a428b1bd0b514a782169d81d03cbec2fed82db67e80f3932e9",
+      "scripts/analyze_loop_kv_redundancy.py": "872e86904e189d18e8440b7f79e80bcbc0061d1b9780aee26c29459710c9d7cd",
+      "scripts/analyze_ouro_selected_wa_kv.py": "f2da633a5e23fa92d9c444b80553149b8644257f479641ef2eec8a9dffd88c3e",
+      "scripts/analyze_paired_screening.py": "777b63be6a0c0b27a07115f94f2da7f50ae8da36818413e6a7b00044895fd79d",
+      "scripts/analyze_predictors_vs_accuracy.py": "0d0c94a94b5c81ab2b11700bbe6a9ceb4d2802ca240d7e1d274954e4ccbee0a0",
+      "scripts/analyze_recurrence_context.py": "2765504af81b1d9025374c1af39d11914977ae1067090292e5a9ecee76a98dc7",
+      "scripts/analyze_run_repeatability.py": "1d529562e36480fa64d32902303c74080f23b5f3358cf04c4e611742df359b46",
+      "scripts/analyze_selective_w4a.py": "ff84fce91a09e11f1d7351a19268086ad9d493e8bdd0657d200fa0128683dff4",
+      "scripts/analyze_sensitivity.py": "5a1d93a6fe3e261f1fe9d54f5535904c378a23b13ef39f482dcea90056c77f0d",
+      "scripts/audit_activation_table10.py": "df38f4deccb5cc437ffe8537a5364277920743c10cdcdede17b1b26709e4e796",
+      "scripts/audit_channel_capacity_reproduction.py": "14558f10129915128726e7535a6f67929e1e71dfc9132865fd94c595ade0b2ad",
+      "scripts/audit_fused_write_model.py": "4bd0788465be0a16fa54709760a5a6ede71d46b1c8128a0f0601bc4bcbb2932d",
+      "scripts/audit_huginn_table5.py": "93913720f049190fa1fcebf93cec2f26d2c802775bb71a5424bd35d227902a6a",
+      "scripts/audit_int2_accuracy_smokes.py": "c0a463d3cd3fea5889c1d758999a3cc059b43c89a57c229513397de6c4521434",
+      "scripts/audit_kv_cost_final.py": "c193155a1d34f2638c05fe65d23b8d68349ceea7586a2acb0f88a8f7cf547abd",
+      "scripts/audit_native_wa_method_screen.py": "05cdf99ca7abbb7333ddce355ca93db4d642b3187fd332007cc5d39bc816d2c0",
+      "scripts/audit_ouro_table5.py": "2966ba1479831d7c941093b5709c00f4b29a7f39a1d8d095025244470cbb35d6",
+      "scripts/audit_portability.py": "465c903e2a1581cc1f84ed7883cecc524295924c31b72dfc685d973460539909",
+      "scripts/audit_quantization_goal.py": "3362e9fc93d2ad3d2aaa89421b8e187c74dd284c6c4a60ad71fe8101505d2086",
+      "scripts/audit_recurrent_speed_goal.py": "7410c511e4ca51c15afedd744b913a2e68fee2b84abab4968e130261ff13bc21",
+      "scripts/audit_scan_g4_model.py": "83acb9495b81dc3ec6d7160c6aceda110826f32ba8f1a50ae1cf5b24d84180cd",
+      "scripts/audit_scan_model.py": "bdd572caf68eef4fd634ce14841512aeb5e2bc64f907cdecbb682d08f0c3a3f3",
+      "scripts/audit_split_solve_full.py": "333caecc946276e9986c9d0a1bb68ab388b4b2b8bbe32b7c1090378cbdd48e33",
+      "scripts/audit_stream_solve_model.py": "11e43ee2ac2f2116684f2818735d980e00613052a06ab53b78993ea92c1cb442",
+      "scripts/benchmark_activation_workspace.py": "1f3c65b83e6b09e211335f0d5710ed9d00533500706a96f36b4bf803cc42c157",
+      "scripts/benchmark_channel_lse_costs.py": "ed77ceb5d1aafc2b1102b874266b6c5bc9e07be8e7ce3e1af14b3a95f0a3fe1f",
+      "scripts/benchmark_channel_lse_policy.py": "362517508242ed94d83e3b11024115be2cd86b71efc7dd4112cbcf0b053a921d",
+      "scripts/benchmark_channel_packed.py": "64be08868ef380c3b905044026db94c289edb1dba849d7aaf74f15d4da6b2d78",
+      "scripts/benchmark_dynamic_lse.py": "0642a3cbb797c802d45bceb60e4b270ea0e793041020320b3489eeff7e14eb71",
+      "scripts/benchmark_dynamic_paged_write.py": "b607ff8de2d0ff083a963cc71d26c7a4d515999e50ebd576353a657d4efbde75",
+      "scripts/benchmark_huginn_native_components.py": "f39093b1d1d339e7fc571db16b2c8c4662b616fd2691a42f35518f16595b4fa8",
+      "scripts/benchmark_kv_cost_probe.py": "6da5ecf0c9767ce784934ee2bee7a0852dd8905a50a921e212520a4d6cfb74c1",
+      "scripts/benchmark_native_activation.py": "ec82e0f1c126b525d49e785f3e085a2ebe769056e77a1a2d13c0aaadc1c755e5",
+      "scripts/benchmark_ouro_native_attention.py": "c5a04ba7798b3631a189187e30bcbfaa08423508f671f08979fa071c445b7364",
+      "scripts/benchmark_ouro_vllm_fixed_generation.py": "92a3d379a73d12350d2c1df1afd98d65146205c412b902bb0051b9a976c4ef9b",
+      "scripts/benchmark_ouro_xqa_seq_trace.py": "d7fa0a4be2e6940e2b0e86071eadbd95004d89b69fe854b05add2fc65cef92b0",
+      "scripts/benchmark_recurrent_kv_chain.py": "085af93b3250eb492823166d7d304ded2ed050dfb42352df80c562c9757f122c",
+      "scripts/benchmark_xqa_candidates.py": "7b130c23368b584b06d3d64ee6dcebd102eff7064ce64edd109bd08563daf4ae",
+      "scripts/benchmark_xqa_costs.py": "ae0b11a31900d21b1e143f5d88fd36887eedcf977154f9a0890870d5fa8fdb6d",
+      "scripts/benchmark_xqa_native_attention.py": "7198b554ed0c10d7d2681c2c297bce03fd085651b552404e4790448d6e8f3cf9",
+      "scripts/build_bf16_anchor_reader.py": "9c072bb017932ae62c2885fcc942c98f8dd7f9e29299151d35ce495e70fc17b1",
+      "scripts/build_completion_manifest.py": "84b8ccc2eafb27025997b981bc9237404cc85ab36f01765101be788eda274369",
+      "scripts/build_fa_tile_attention.py": "5ff13df779225048a1b8f42de46e6a99993569edd51725ab2e3067f0f1bd6ccd",
+      "scripts/build_m1_storage_matrix.py": "a5eeaa9eaeca126e03be047ee4537478f068f5a730616bee8a36558a9f41be86",
+      "scripts/build_provenance_manifest.py": "96eeb4ec1c23af52a5e4ffb415aac532356eb37370c862d5176a4053c100646a",
+      "scripts/build_recurrent_attention.py": "0896c6280513fe2b624c16451e8773007e2acf685adbafb467f89a9127c7f950",
+      "scripts/build_recurrent_read_view.py": "def6ccd89fafa3db068972e426c348e9d29c67d2a4f3506529e3b44c0509d981",
+      "scripts/build_recurrent_xqa.py": "a87ce431cd27ac95476ca6382677fb28093f5942a4a87ab9c7f2d057adbb0a9f",
+      "scripts/build_vector_decode.py": "2eba31bc992c7895b1c17b8588590001617e9707ae2ffc8621f82e572b56369c",
+      "scripts/build_xqa_candidates.py": "4e0e4cab2c5927025f5946a5a79f73f43c3386012ed5b4b95741cd63eb0c7ffd",
+      "scripts/build_xqa_comparison.py": "8fbdad07ddd5ac35df2acf14a5a01a8522eea83877b7cb35ec6a553cd72bccb4",
+      "scripts/build_xqa_cost_variants.py": "88d90a75f931a205dcbb13079846aba305a542e58b0447242194ee39c8b8b5a3",
+      "scripts/build_xqa_hnd_native.py": "a3e0ed25569036b194b0fcfb6f59b8478145396e351cbb0db05dc83ffa2b48c2",
+      "scripts/calibrate_huginn_residual_activation.py": "7189b032c99b5f5c25d2349481cff3f97a4a919ad9510c8e925f97315c7c25b2",
+      "scripts/calibrate_huginn_residual_kv.py": "a5809dae0f75d5c7e45b2453ff1576e3bbf05c1a86f6fcb17fcca03bb69da0c0",
+      "scripts/calibrate_native_recurrent_kv.py": "de867b19016e5cf3d90fac9b1684f13874bebdcb7658e34aff1a059403a0cca3",
+      "scripts/calibrate_ouro_rotatekv.py": "f995a50d5b35adda2560c1f26ea44482cd78145968a07336fbd06868ce50e553",
+      "scripts/calibrate_residual_activation.py": "0c5348f5a0989bf4f178c877b637dae3d9a3a1225e28986760aca39002c5a600",
+      "scripts/calibrate_residual_kv.py": "1b7d6c1c23bbd251dcb7ce3f9f1dd994b581e6c8d8237a0f7c9fa79c950ff9d2",
+      "scripts/capture_int_scale_matrix.py": "f507a18d3394d0b39e94bcbbb45e27cfedb68a9435a3b9fe27079cc365065bc1",
+      "scripts/capture_ouro_kv_reuse.py": "72f411ddcc7e546874aea70e54cee946ddce7e6f85c558087dfe1e2fe7a940da",
+      "scripts/check_activation_workspace_cuda_graph.py": "e0d2c9b871949286a6b0c45b8e5a8b713c6f8f0ad2dbfbfac033084308be2310",
+      "scripts/check_bf16_anchor_reader.py": "107caee9b7a3a58d1554016feca042c545240ce5c4f2685f5edce0d1d783c6d0",
+      "scripts/check_channel_lse_certified.py": "141e493ef9ac943fd65e4b0352c48844ab1e250ee577064d98b4790364ca93c5",
+      "scripts/check_channel_lse_native.py": "7937d8c0d3b94830984ee8fa19c9af7829689ce4ff1ca20f448276d587fe993a",
+      "scripts/check_channel_lse_split_solve.py": "ed222515a14b35c617f80259d91d53b98563a2928009f6126d7a7b93844b1734",
+      "scripts/check_channel_lse_split_state.py": "8634d6c172652760f57070b2ae9a782b3c1fb865832744591f3b89ee2272790b",
+      "scripts/check_channel_mixedstats.py": "eb80353d62f2569b012589290510a165d4a9d04190f2e15c319534c366401b32",
+      "scripts/check_channel_packed.py": "7d5983fd3850838f6da437196f29459f91248cf223900c063f5032ed37195094",
+      "scripts/check_channel_packed_heads.py": "4cbd129906f2ce7a5f4e91eafb50724bf49eda9df4eac126531fbdb80b342c1a",
+      "scripts/check_channel_packed_split_solve.py": "75900d0a014901182629864c5575ee0d01bfe4e1b92b3dc387535b26560973fa",
+      "scripts/check_channel_paged_fused.py": "72bbadb9214c5815050f1c3436659a6619632c42d6bcf10d4cadca8665a7719d",
+      "scripts/check_channel_request_reset.py": "eb785b1301ce8380eb7052b4433c63891f7d4bd14a369125af1930ce2a5ae4eb",
+      "scripts/check_channel_scan_numeric.py": "11c090307331ed7078c0986e3c5436bf1b5387f078afe78320bb9e645d5d429c",
+      "scripts/check_comparison_attention.py": "fb8b20d36f57efd3eae00db870471e285eafcc68bfaf91ab9a7a54d119aa0571",
+      "scripts/check_dynamic_lse_recurrence.py": "7d6ed664c3004ff6445ccf9f6d3d7c4501fe346a6e5b13a745e159c0855e669f",
+      "scripts/check_dynamic_native_scalar.py": "fcefa1178f06cd7281c9fb67c858bf70b6857afce5699b3617f11b89a97bdda7",
+      "scripts/check_dynamic_paged.py": "b42960a5f7795f13646e98acc937b810b9f1e12cc1b9ba0d8a0d00a3feae12bc",
+      "scripts/check_dynamic_paged_fused.py": "cea6ff572d72bbfb4f03a5b0325eeb6bf17758c58f052487d4b40c3eec1ba456",
+      "scripts/check_dynamic_paged_heads.py": "c17a61e56ba883cb92b9aa54a5d8b557abd337af32cf9711b90979f7dc0b042a",
+      "scripts/check_environment.py": "00343fbd026557b2c317c26883c37d0218495512cb38bb99506ecb1c744d5490",
+      "scripts/check_fa_tile_attention.py": "1b23aa2a0d25c3176c175c8e841e616e112ff40e70a4c1e4178ca881e629d9e2",
+      "scripts/check_fp8_read_view.py": "5d2cf0e70bde90177c5a6be7234c6bb94fe303957e595ff2f7ea1858cab49708",
+      "scripts/check_huginn_graph_parity.py": "e1b5002ed78f7ba49a23e62db86f29841365265330c5d5cf3bcfcc87391d8695",
+      "scripts/check_huginn_sm120_fp4_parity.py": "6916213489c904c13994ae663b7f219c3890d45c9247353600e66929d3c3462e",
+      "scripts/check_int2_experiments_gpu.py": "400ceb0731884a208d3b5225fb4994e682a92a0163ec854a308c55c9fd62b091",
+      "scripts/check_int2_streaming_gpu.py": "dbaeb741a1215f06de00b78bc0f71c6e6c01b8a100ad9c180b4ca0f6848c2b5c",
+      "scripts/check_math_modelopt.py": "7924917242ed769f80ce00d18043ef7b719706c480259182c076f98ba270b62b",
+      "scripts/check_packed_affine_kv.py": "acc10f3ddaa6b38957c42f0741301668700ad2882711117fa990d574feb53b20",
+      "scripts/check_packed_fp8_conversion.py": "b05e9ef6f321715f93341d079c4a0c13cd1c516763d91be4b55957ca02016314",
+      "scripts/check_paged_comparison_attention.py": "c35a2f8983149403f6f13986ad38e4e3df331908cbee5ffafb70eeca75268359",
+      "scripts/check_recurrent_attention.py": "18d513c8b2545f7c0f393b5edbdff0451fef22bebb6f2a3b6ca7c90083090179",
+      "scripts/check_recurrent_dequant.py": "3717d5bb5d012a73e3c51416330d82310e6baa81c7777a32277920a78b890e2b",
+      "scripts/check_static_paged_fused.py": "69d122d8d6bd49728990ea71a5437a1ddca2ce8a2d6e01e3e15cf067f1372e55",
+      "scripts/check_xqa_bf16_matched.py": "34a5a6469befd2f372caad2b278d1ff1d9b0b06b842e63003c36c81927acbe40",
+      "scripts/check_xqa_fp4_lookup.py": "5aab5b1a773ef3495e65e69cbeb36657d884b1cd0567c831285c91f747edffe4",
+      "scripts/check_xqa_fused_prepare.py": "2fb54c3928dc77e369887517396bad325ea44ec019c2ec29bca9c0f85925831b",
+      "scripts/check_xqa_half_product.py": "b55b1632d22a562271815e55250f8f6ca1493ae8251f1fcfbe56e14d1eb8e61b",
+      "scripts/check_xqa_mixed_attention.py": "b0708b1cf369377d5e65efd637b2126bef1e33d5be7940e0b2d7fd6f990a1de4",
+      "scripts/check_xqa_paged_attention.py": "9d0fad47bb905da01d8c0eb2d76084dedac321bf9771fc349de6ec42b346d870",
+      "scripts/check_xqa_paged_generic.py": "f3efb933095fd47da14c64e86d463dbfeb4e7731c5e0dd50cec2a6bc8f2ae15c",
+      "scripts/check_xqa_prefill_decode.py": "bcb51bd5947ccc1ca2c490ca8a17dcfdbf1c68312a6a441b724ddfd1998184db",
+      "scripts/collect_activation_momentum_payload_heatmaps.py": "e773e670b76f5b5a0af600edc340535f028121d122daf2bb3d5c50c13aaca2d0",
+      "scripts/collect_gradient_tensor_norms.py": "a1d6d20ae81e700badf807e445a770a24a2f2dd2cb50c7659dc07efe910fa031",
+      "scripts/collect_huginn_anchor_mechanism_cell.py": "b9c58fd7fc3fd6caa8ee5875f4a8619ec736646afbafcc4156285e15d0caa055",
+      "scripts/collect_huginn_layer_channel_heatmaps.py": "87977e4d6a34e3422fa42af1ac943cbd17814eb6911f8f19297318b72ddeeb6a",
+      "scripts/collect_layer_channel_heatmaps.py": "41a293cede5636d07ab52baa61d7c495b63cfde5915ad9daa8a15db62aea99d7",
+      "scripts/collect_momentum_payload_heatmap.py": "0b94a57d71fb7b8318c21f9f2e6fc2e604317debd441a1a73fc5d0dd76c28c9e",
+      "scripts/collect_ouro_gradients.py": "d7b7c0671906c75c353ff34b257b60ccd5405479647f85bdf957204834fbf493",
+      "scripts/continue_after_q08_uniform.py": "10c0b22748aa4066bd69c49f9dc4e9c4ae50068fef2e1bd7b953bce4568412eb",
+      "scripts/diagnose_channel_prefix_cost.py": "f29704a8cc5d3fd3ce81892fcb06395349d15ed64a32d41f2936be8a18121e6c",
+      "scripts/diagnose_int8_scale_underflow.py": "a5c134d657da16dde0cab5eeb09fcbd3e3c2b837e8ea8447defe0354f6546b40",
+      "scripts/evaluate_gsm8k_cache_codecs.py": "2bf64cde6944cc6cf3ac3e88ef8d22336a27fc4da6c2d7b4505c0fd930a275cf",
+      "scripts/evaluate_gsm8k_fake_quant.py": "22d0c237f74310ed2a8de0973b8bbbe59bee5c57ade8e5fa35f51b7e941d0b8b",
+      "scripts/evaluate_ouro_attention_precision.py": "8fc0f383cae171c45fb3029b83ff42bdf0e0e37b761c3e6e58abb4dd30ff9562",
+      "scripts/evaluate_ouro_code.py": "649415991ebaa7894c40e0900c76408ee2fbe27a6a483f5438cbe2d1d99712ff",
+      "scripts/evaluate_ouro_recurrent_vllm.py": "d0f14e137f425dc1e0f6d2341dbefab3f4df2d04a313760ae99e2b2ff80bee2b",
+      "scripts/evaluate_ouro_vllm_attention_precision.py": "b945f987d6dcc98a25f8fd2f1eb307774e3d4b6cb1170c5e2b2d981fa54b9ba9",
+      "scripts/experiment_manifest_common.py": "49b5ca26f2087394144ddaa6d0a36b07f16e503b7a838a382fd7b8f2257b74e7",
+      "scripts/experiment_queue.py": "21df177893db181b87b13dfc04d9e5538e81cb4770cec8225df7485d91c0badf",
+      "scripts/export_codex_sessions.py": "212d5d0428d31aca43345cf6017ebce23dbfa98134ac85ab0b1a8ed16f2fb0b0",
+      "scripts/export_native_nvfp4_checkpoint.py": "b7373493e7a13586827c370f8521e544dcf238a16792c89ac60174c8a7eb77e9",
+      "scripts/finalize_activation_table10_report.py": "4f9bb32ce670eb61f270e9dfcfffb7786625c5e73c42b111e0c66d792a1eff41",
+      "scripts/finalize_huginn_table5_report.py": "dc1994ae56f9100a92c4bc5417c10bab9b7d148d81200732740acb2522ad17aa",
+      "scripts/finalize_loop1_dynamic_table15.py": "753e956cb404d2401cfca8928eaf5d8cd568a9d17c1b93971a10a5ba2e83123c",
+      "scripts/finalize_loop1_remaining_resumed.py": "94cdb86481d84ddcda7abcc7831f9b3a93da857cc41cf3e592453c8369fa224c",
+      "scripts/finalize_q08_uniform_bits.py": "3c653a6ac63e6495e1c293fdfaee85c302cd657d770457e0c63735cc5910c752",
+      "scripts/fit_int2_accuracy_candidates.py": "bb9d87115ac4213cff11d12e96d83e1e5c4eade0ac67821fb5b3df326c9361d7",
+      "scripts/generate_activation_table10_tex.py": "f8f855ba9870dafab119fe9ee3da4d085814f36270c1286296a2ea4d339a00ca",
+      "scripts/generate_huginn_table5_tex.py": "7253a086a2cc10630211307fec2142c7e12ec06e010e0c64662acdc0a845f11e",
+      "scripts/generate_kv_codec_report_assets.py": "9f0fcfd69147f1bc7feb3a54b9cbaa0c4465bd73eeb92b4d8f8fc99008dbb975",
+      "scripts/generate_notion_goal_results.py": "c24ce41bf220cef7f3bbcfe3b0b656a9fce58c28a444f71506f741e683ecbb3e",
+      "scripts/generate_nvfp4_residual_report_tables.py": "830a8a6c038f739f3dc44c056601db515fe82f3df94c118dbf625dfc190d8dc0",
+      "scripts/generate_precision_flow.py": "24b713b71784ec3c84f52940b535d1027bd5ca40fb73f81baaedbffdcf7a1bda",
+      "scripts/gpu_availability.py": "09c716b2fad2e969b3eda9acedf91817d4dfbf0d67280563bfd2588386a486d3",
+      "scripts/huginn_table5_matrix.py": "22eb8f679a832ecbbe6f2d43ddb9653c2103ec3417cbd246b56e33abd0e59325",
+      "scripts/measure_sinq_nvfp4_weights.py": "cc2b80b416b5c47b415bb4c60ece87d394ab61b751fb726c206abee9c7c11d58",
+      "scripts/merge_accuracy_results.py": "a8d0d79613893024382ed874b497fa08154bd1ffca88e3bf8ff669216ef378a0",
+      "scripts/merge_full_validation.py": "ea7182148e8b98defc87b664176570a75ab08f9bec5c05799e70c25b635cbf53",
+      "scripts/merge_gradient_tensor_norms.py": "850ecd33e3fe3539fc83b6adb8232d58a8e3f6a540ec1c07c4cc17ab6b38205d",
+      "scripts/merge_huginn_anchor_mechanisms.py": "8d02dce1f8914cfcfe18118378dd10c370d52b619e598bc181fb65d6e7b431fe",
+      "scripts/merge_kv_codec_full_validation.py": "579713ce35513f10537b2c3abba3f7b33aedc05d73aa76c01d103c85adae48cf",
+      "scripts/merge_layer_channel_heatmaps.py": "5a56b2407d5ada4ab61415e00f25c4780926d1ce2bffd03707576eda52d6cfd9",
+      "scripts/merge_sensitivity_shards.py": "1b1afa38007afe6bf4d4e54e211cbb956b3b6af79b1fe0a24300a33c99851e06",
+      "scripts/plot_delta_ce_summary_grids.py": "7029fe408fc3e2e8dd29407d22c2c6b8fa574b8a374827791704c71271282672",
+      "scripts/plot_gradient_tensor_norm_heatmap.py": "f888343b48a10cf142532c391768526cae058bce7a70197453dc1387715b0fc3",
+      "scripts/plot_gradient_tensor_norms_v2.py": "7a057b1c2b3a7251eb8ea565a64a32a82cad4bde00abd2b4c6ac53bb0d1ae00c",
+      "scripts/plot_layer_channel_delta_ce.py": "c7167a3a26b92e7804c4594494e0e102df57fb3293efd15a703d97e37e9c3c9b",
+      "scripts/plot_layer_channel_heatmap_grids.py": "f0ccfeb3f0dbaa6e3e7710b2e8e36a5dbded48565aeae1dd78565c1fafdb4d0c",
+      "scripts/plot_layer_channel_heatmaps.py": "e436e4e3b24945bac35785d0c2882ae4ef253d9ccb88281333230ba0275ef601",
+      "scripts/plot_quantized_payload_heatmaps.py": "8b652720d49a64b32336f09b3b7d72d21b754d6c84d238a821c612ea9328d37b",
+      "scripts/plot_quantized_site_gradient_heatmaps.py": "89859986a1625c6143a562eb6441c99760c7a310f287f95cf82aed0551a72ab3",
+      "scripts/prepare_huggingface_release.py": "a0ad51ea23de34f4c5023d61eb6afe23c9ac65a4c1d151c7ad574ff8c3614326",
+      "scripts/prepare_huginn_pre_sinq_checkpoint.py": "2ea485abd7d3f9458e243a768d8e4fd54ccd0bf23b2b98489608703533a1bd2d",
+      "scripts/prepare_ouro_pre_sinq_checkpoint.py": "923838b1d95298e9ccaf81d11e4dd39c6f4a7834596e354075f7d5ee60253be4",
+      "scripts/probe_split_solve_vllm.py": "e8d8b3ac144dd8c9b8574f3c0db35f706a1f97a1153bc3b7bacda4a14fe04f8b",
+      "scripts/reproduce.py": "7479a7e865d62d16d245745efc964e532eca419ded3aafda7d2a4951cec88697",
+      "scripts/rotation_study/after_hnd_memcheck.py": "bbc9751a541c33a3f96bc2d3b5d14657e924b3207ba96a21613ffa69b9797c11",
+      "scripts/rotation_study/after_microbenchmark.py": "e0ca715d17ffb68bb61671a19250d546b2165be15ca903db70a2448b8b1685b4",
+      "scripts/rotation_study/after_resource_audit.py": "06f2c32b4f0cdc6ac7dc91e1f7b529c731cb3695af97459b9171326898b5d8a9",
+      "scripts/rotation_study/assemble.py": "59f0f0fe62e17f54b63deef52bc8c6db3173d05d781632a055f58f5f2271e642",
+      "scripts/rotation_study/audit_adaptive_safe_resources.py": "e9c9a60429e83b33391aa9451ca452ed2a0b0f10b998c498ea580903c2f47256",
+      "scripts/rotation_study/audit_combined2_resources.py": "949ea01ea47dada23e561352ad10fdbca1bcad40febe89043b39a73e6d9f9e10",
+      "scripts/rotation_study/audit_combined4_resources.py": "5974bf8fe837f653f01e72aaeb19e5a5b3d5da82e3e131db136e5e1fc699b0bd",
+      "scripts/rotation_study/audit_extended_resources.py": "23522c6deab12c048be5398e26d13e0286dd58d337165e4e82bf129d5883e1c5",
+      "scripts/rotation_study/audit_extended_resources_v2.py": "c1d7d7d2dc185726c60a5fddfa54e98da9ac59fbd146a7956e03a323c79a14c4",
+      "scripts/rotation_study/audit_kernel_resources.py": "0e75d8bc8c926167f2bbb80b929aa2a0ed1987652757ea5735062411b11f10e5",
+      "scripts/rotation_study/benchmark_attention_conservative.py": "0738cf41aad38302ccef140f84a5ff4c7369811f77ce10f8a65dbf538d7f6a54",
+      "scripts/rotation_study/benchmark_attention_fast.py": "7ab5657d2d65600f4123764fc2fa68710a5bcee7ea9d48315a159f582166181a",
+      "scripts/rotation_study/benchmark_attention_group_load.py": "7731b1c4ba41a4c1233d9affde6ff390adc52e8694cba34d58191e6d90c2ed67",
+      "scripts/rotation_study/benchmark_attention_hadamard.py": "c9dfcf998b9966f24be6bb28beca5860b1c5f3e3eef2d39c13c7aaa17e08b76a",
+      "scripts/rotation_study/benchmark_attention_pipeline2.py": "72ade1025ea6b7a7cd75c16b9dc511288c0457af1f17650ed9a28483bd7c568f",
+      "scripts/rotation_study/benchmark_attention_query16.py": "50ec0df96debf7efcb79be22794d9224a41ef8a6661ed1934fcf700dbcc9e4dd",
+      "scripts/rotation_study/benchmark_attention_query8.py": "b694221a3443dc5678eef6fee96001451d1b0b6b8a8dec4c13d6353c23459e9a",
+      "scripts/rotation_study/benchmark_attention_unroll2.py": "a33454010fd4476c29668f9b8244303dffc6e1f809c4fd1d84b8648d0b944058",
+      "scripts/rotation_study/benchmark_attention_unroll4.py": "75aeba671bd83fbb51b9ccb53afffe3cd5a33522653ddc488974de44ebc1d5b4",
+      "scripts/rotation_study/benchmark_attention_warps4.py": "b9b5ae3e8e6e99c6749624da82a13cec9242289438f6fd9cb2f60f71ba050f95",
+      "scripts/rotation_study/calibrate.py": "229bd2299d8d2fc581c9070788e63b75b991e15fb1d713952fecde6a3f888dad",
+      "scripts/rotation_study/check_attention_combined.py": "bfdaa9f9f05064053372ae3cecd3b111817676ddea88ae4d7a50a46adb74b9fc",
+      "scripts/rotation_study/check_attention_combined2.py": "102fc9d461c92391debc06443f045c863f935ef3526b0edb8d99c413c25ae17a",
+      "scripts/rotation_study/check_attention_conservative.py": "3c618c8acf3761c91fc03162f76209f92875b909bad12f9d240892150eecef33",
+      "scripts/rotation_study/check_attention_fast.py": "960716ba5ccc9fea0daf7c2b5b7807862a30a1ca6190096969b4ec98cba88a88",
+      "scripts/rotation_study/check_attention_graph_growth.py": "e13e721cf4c77e319832f079d37d84c0499cb8096f1a8885a2dd4224b2b5cf3b",
+      "scripts/rotation_study/check_attention_group_all.py": "903fb7ce4ef652b9a83b3f7dc670b7eccd81a4c4c27db4c036e7f35f7c561df8",
+      "scripts/rotation_study/check_attention_group_codes_v2.py": "1c7adde1dfa15bda45403f9e3280c97bc41c2917c41cb3a4b09e1e8b57bab1f3",
+      "scripts/rotation_study/check_attention_group_load.py": "e0df7bb939e31299536af8711816defb19e4d6c0ed9922d9ad3d5338c0cbb6dd",
+      "scripts/rotation_study/check_attention_query16.py": "7234778ff9fcfa4682cb8c473f2aeda2596cf05564b5c5108d9a0f8028a5a058",
+      "scripts/rotation_study/check_attention_query8.py": "67cc348b035731dce563728a7c502d980ebff9c218ec87b06c8a082921f1415a",
+      "scripts/rotation_study/check_attention_xqa.py": "f1ea76d15c447ae926daef4532c86de9f74d14c9e2c7ec6ab731439869ef20e1",
+      "scripts/rotation_study/check_clipping.py": "456503eb2ceb13231f896e6bca8bae707d0fb8c0c34f3bfbaa88813089044766",
+      "scripts/rotation_study/check_packed.py": "75ce77cc5689f0d580418631b1b7d651db88d67cc2214af80b56f1d6cf839bd4",
+      "scripts/rotation_study/check_packed_extended.py": "a61f4df471eb030553b5c3845773743ca4ca90e956550d4267bac289aa8f70a9",
+      "scripts/rotation_study/check_reference.py": "96a1fa527711697f34ad9f9b9721da53e842cde85f9799c64203a022721fd328",
+      "scripts/rotation_study/collect.py": "175fb6af33c49bd09ef5df3a6329aebc97a2666c5b605442125cc930f1eed766",
+      "scripts/rotation_study/diagnose_hnd_alignment.py": "f49d2ff32017adeb6ed60bf718f7397e74e83bea0d15b5f064fb8d58112f3339",
+      "scripts/rotation_study/diagnose_hnd_alignment_safe.py": "f7c3e64765b0cc6b86b4e374e46485c92d20fef33bd6a28984bb1a7e1172d7db",
+      "scripts/rotation_study/evaluate.py": "32621676094335cf3d96eb81801d72288d9d6390a7329f9acf9c06a3aabf3a01",
+      "scripts/rotation_study/evaluate_combined.py": "1d755d82819038b94c3376ae6f9c06b6cc8e33119f8aff3722ea4de9bc8fd973",
+      "scripts/rotation_study/evaluate_combined2.py": "683300fe00fd9a2bef03c7bce257ef7a290e032e901b5bc90fcc5fc92f1f45ab",
+      "scripts/rotation_study/evaluate_conservative.py": "fbcbb6cc82cbfe1e16515b3f5f068ab356ad214ee37bfbab7adca15da39eccf7",
+      "scripts/rotation_study/evaluate_fast.py": "4a6ccfac980e651f6813deffcd8ea14d435186cf29c4386e3ba4041b169f434d",
+      "scripts/rotation_study/evaluate_group_all.py": "ed9a980dfd7b2140107ba2d0748940df49cf636af0d85abb804b82838c946216",
+      "scripts/rotation_study/evaluate_group_codes_v2.py": "e5e24323924dd11083681d5a9f74ceec40677463ba9704ab8d479554dbf530fb",
+      "scripts/rotation_study/evaluate_group_load.py": "6b5f72d640017a8787d4f4c00d4a155158b662e073cc68f3585de7a8a6d35d25",
+      "scripts/rotation_study/evaluate_query16.py": "cb3d9282444f2c2a388ef16a7875a88c69c88f8a4bd4314bd91212e2124970ec",
+      "scripts/rotation_study/evaluate_query8.py": "b14553e09b497cbc9b64b7f12ee1d8770ae1db01d1efadc1f1afd0ba39810e30",
+      "scripts/rotation_study/evaluate_xqa.py": "447268526c0b2a146a59a0ebfb81fefa5e7120050322f95e2ca212ac1bca3074",
+      "scripts/rotation_study/finalize_recovered_study.py": "0659a2c8cdf10e7f297356a6f3cae744f63c6f6494fe9167314946c2c77470e5",
+      "scripts/rotation_study/finalize_when_done.py": "0f97ba9270afb4782d699815c0afca65e462a363fd0f626f0733bc8a81287e90",
+      "scripts/rotation_study/pipeline.py": "3e79d4d7fbea491ea361def4ce86c2077871151bc8ac8b4d2c44664ac3eaad14",
+      "scripts/rotation_study/profile_attention_costs.py": "91ec37ff0c5e316c68aa9f449f696da41e9e1facebad4a6e03f61954b7e4d6fa",
+      "scripts/rotation_study/resume_full_cells.py": "fe0a80d87a046343119fe3fc0771ba8597235d1879ce8ff0f60102fd1f87bd42",
+      "scripts/rotation_study/run_acceleration_comparison.py": "2edb6c1f4fb5d6d7900b0a26ab3e4a1890f8fb030527648952299b8cc62d5771",
+      "scripts/rotation_study/run_combined2_comparison.py": "f344b81b7e96f7db40a49a2f8f52da685f699e9005f43df78da7a5a2a1d1be94",
+      "scripts/rotation_study/run_combined2_followup.py": "d403d837d6d62d4d6c50fc0c3e06ecc28c273bc0842966c3a312dbdf38915cbb",
+      "scripts/rotation_study/run_combined_comparison.py": "a4a276513fa311a44488d0fdade035916d5a8a238ab41a619d7152b8a6585c42",
+      "scripts/rotation_study/run_combined_validation.py": "c8d22d1f36189ba7d8b77b2284b5bc1d1b71815d4496ac0e4b92eb1681db5bd1",
+      "scripts/rotation_study/run_conservative_comparison.py": "61405a1b15adc4d3c0e0c7db08897b98300d7cacd7183fbbb34a8da63db25412",
+      "scripts/rotation_study/run_expanded_acceleration.py": "e89890efc49cbc49f544c80e6e82fcc1acca7032bdad038a5e466e9ce8fa0f1e",
+      "scripts/rotation_study/run_group_all_comparison.py": "6dd53bc538e3b377535ef83bbc344e64db37a3d7ebcb4cb6e276f9f1a6251186",
+      "scripts/rotation_study/run_group_all_followup.py": "e42d4e2230db6a02f589d0d8a5d9dfdb09a24a2c5d49cc0a063f88407730a3bb",
+      "scripts/rotation_study/run_group_codes_v2_comparison.py": "129de34774e3191999e2c75fea8990531b4a8ae9ed8392d935710ac18214e177",
+      "scripts/rotation_study/run_group_codes_v2_followup.py": "457030747991052eb77fb3152cbdca78584abdcbbd1e5cfff98d22450c0090ff",
+      "scripts/rotation_study/run_group_load_comparison.py": "61e770792b1eb50f5bc76cc2f8368611839fe51353e84b89366b07935fa215d0",
+      "scripts/rotation_study/run_group_load_followup.py": "c23dfc84017bbe3d7f84e418bb51e505335d732907327a1ccf2a44f9e1387af7",
+      "scripts/rotation_study/run_loader_screen_queue.py": "825bbda7fbc6549ccf470ca958349110c0487d43a752871cd70a4a563dab710e",
+      "scripts/rotation_study/run_query16_comparison.py": "ffcaac8aad3c9db92567e4bef208a484dfb4063781f8732d3ae16f7c581ee9f9",
+      "scripts/rotation_study/run_query16_followup.py": "e8a7efda0d5a50e7e87d939e90cb755485ef4975a2295a4710157ace4284ffab",
+      "scripts/rotation_study/run_query8_comparison.py": "594aecbcf665813567d0f5c61c675504da6cd77f241f597f856f5f60e500516d",
+      "scripts/rotation_study/run_xqa_comparison.py": "ecdb552092650d2cba17a36e8fbe736d67c5a08737f15183ee3faf5f55ec942f",
+      "scripts/rotation_study/screen_adaptive.py": "c47cb8699940ab6ed5c822ec203d73545dfd290d1bbf01d9657ff8cf9ecd0a79",
+      "scripts/rotation_study/screen_adaptive_safe.py": "d40993ff4e40a9a617f54c35165ef52c6d3c58146e8117f112992b3433aa469a",
+      "scripts/rotation_study/screen_combined.py": "bbca232a238a01f9bf54252723980dbbe0c132d684e551334262be4f956f9fae",
+      "scripts/rotation_study/screen_combined2.py": "b35609da38f021ce247f7cab965e8dfa0ea7a2be9abaaa1c8175cd9e1f5bb840",
+      "scripts/rotation_study/screen_combined4.py": "7912e85e5dd2cd10ba2b87e4099ec5d50e1b7229f0ad024f0fe3f5afe6005b59",
+      "scripts/rotation_study/screen_group_prefill.py": "8b50b8a773174e37a1467db8cc7cb9d25e9263f8fc0239ccdc4097c8f8cb360e",
+      "scripts/rotation_study/screen_hnd_cache.py": "d208cfcfd036325d7d1a8b406480fc1f432f3833172b26426bbf124d0d34b6f6",
+      "scripts/rotation_study/screen_hnd_cache_safe.py": "f7251a4dbffeda1f465bccbd48f9ee201e10d461edb0998c748abdc596752eb3",
+      "scripts/rotation_study/screen_loader_candidates.py": "d201f41d4bc5f5c5f33f2fc62c601087f2eef629f1e1e3e4a87a4732891ed3b8",
+      "scripts/rotation_study/screen_loader_candidates_v2.py": "c2aee9b56fc05510f38880f2072a12b65efc938d67d5fe391cdd6a9a8277edf6",
+      "scripts/rotation_study/screen_rotation_access.py": "5cd4cbd0df0aaf9d7c8e154bd4a206ed5f1f9b47155659dafaa87641d4d71a54",
+      "scripts/rotation_study/summarize.py": "54be9e99168c4cf430cb27ba10de54fb6d6e35268b4466c9b45b0ee63fd246b3",
+      "scripts/rotation_study/summarize_expanded_acceleration.py": "1d15778f3b1e093ebfd8b4deb8fccbbe99035752ca691f2b0d0be445a28ac933",
+      "scripts/rotation_study/validate_acceleration_finalist.py": "1432e674519f118fb26ead5ba5d1396b80b3cb42e4f55b95be6d90a4c6a389a6",
+      "scripts/rotation_study/wait_validation.py": "26265d2d904c2734be7fdc2a518ede8b51493d22d7785b332d32c0d77f383aa5",
+      "scripts/rotation_table1/assemble.py": "85022a1d7ebd6758206162ef7eb0420b580e2c33c826db908892e8198231b680",
+      "scripts/rotation_table1/assemble_identity.py": "07c311f0589b47010f958add0defbc9b789af481d92e99098259cedd744bf528",
+      "scripts/rotation_table1/audit_smokes.py": "537a099a365dc36f616059887e280d369c363d4088aaec88780e77755e0fa412",
+      "scripts/rotation_table1/calibrate.py": "bbf40471f159f7af193362271aa4d2a7b6ce34e8a32959abc8fc7983edfb0b1d",
+      "scripts/rotation_table1/check_math.py": "ff0b97813f57eaec01035859d178c5b8aeb9828ae34008acd32731fc363e7fec",
+      "scripts/rotation_table1/check_packed.py": "57aa4980c083806bed498a2025f3e4f868cb03a6e35867f4cc722d4dd7bde961",
+      "scripts/rotation_table1/evaluate.py": "37b81a409d5bf8b9f5449839022d98059fb90c851c19c7a19c32f4599a7ae902",
+      "scripts/rotation_table1/plan_table3.py": "d6f579cab190f6442f5d6fd1b74f61d5244a1a78698eeaa76e120e01b5267212",
+      "scripts/rotation_table1/render_table1.py": "eb9d6e548fea8f753f11475e80146b4cc7b535a76a24c2d67bca9fdca42660b3",
+      "scripts/rotation_table1/render_table3.py": "5bc33b107c92760cfd6c3dae3b0dd7e8e72cb1e68533b17a44208c82081b40ab",
+      "scripts/rotation_table1/run_full.py": "b4bdcf7fb41d4e79525bd146a0fc589792f721769cb6441e7a17a799d4c475e2",
+      "scripts/rotation_table1/run_smokes.py": "d216bf8354b3c9d18bc90ba103a9095a39b80c8c20acab27656ce1d7828fbf5b",
+      "scripts/rotation_table1/run_table3.py": "1ee705fa75413c69fd0ee11a44029cf2c1bb511c36cc9858f2cd65bb35d1f258",
+      "scripts/rotation_table1/select_method.py": "c1d2f45d1f7bff4639e9c12211889076289483176a84435e3a9dadc7a46f572b",
+      "scripts/rotation_table1/summarize_calibration.py": "4b5bb9bd7d07cba440da7ce89bf365de32ee8b6d2843aa23dc4969bbb85d479c",
+      "scripts/rotation_table1/summarize_full.py": "bf94aac5c5ccc07cbfd82e38f01834e6989f386301f45fc8f378e177f688f81a",
+      "scripts/rotation_table1/summarize_table3.py": "1323e4a7cb40fc13862d1b455b241f241608cb151b72dfa629363f540cd843f3",
+      "scripts/rotation_table1/table3_jobs.py": "c368ef161c9bef456a39d215673f7b0f98ae75cddc8363cbf1b3ce1dcf533faa",
+      "scripts/run_activation_predictor_matrix.py": "5e9756e04612221cea145990d118443c3e747d516bd1ee3ce21d4e0403ca763b",
+      "scripts/run_activation_table10.py": "be359602b5e147ddd0d28a8f28726dbd9b857622bcc037bc7843e91caee992bb",
+      "scripts/run_coarse_sensitivity.py": "f327fbb2e2b80cb98d6b00b49e96484dbce1f479bc67d4e33b1b6cc9ff2d16f8",
+      "scripts/run_dynamic_group_scale_multi.py": "69d91faf446b9103d4c00f5175ec1a0f786b5e719bcaf3c9f94872dc87bf01ed",
+      "scripts/run_dynamic_group_scale_sweep.py": "4e8588a8f1ca7f02b566d8ede5373a18cf56a014d84487d458bcf4b7c2d64b68",
+      "scripts/run_fixed4_suite.py": "46eacb9ddfa89466cff33a1610d2051e53dd22a2489c933008757c1ba3eb8605",
+      "scripts/run_gsm8k.py": "594e570e58ea79aaaf27e6fa20088e4a63856044cf292b3d8aec57e658ea31ae",
+      "scripts/run_huginn_anchor_ablation.py": "d249e21f49f7df8421006fb847b51ad4848725ec0b7c16f4e2601bfb3599793e",
+      "scripts/run_huginn_anchor_ablation_queue.py": "544e241d06cb2aa34ab7a5946c453ef184f64fda5567689989b2a2537dfb3b65",
+      "scripts/run_huginn_anchor_calibrations.py": "fd4f8909b3fdc08745a5f26021096f52999fbed7cf510f06a6c4135acb0907ce",
+      "scripts/run_huginn_anchor_mechanism_queue.py": "b201ca02b2f18849bba4c05103fe5323d428cae4bbd919c8db268c1bcee4847e",
+      "scripts/run_huginn_full_finalists.py": "c6d79cc94e4d2314c7499f785cc196ed2f31cbfdf12d19ed18fcad8704bdb941",
+      "scripts/run_huginn_heatmap_queue.py": "8958642e837aecd51f0eb1f5d05b873974983ca24fd02779e7a03cf94a914a34",
+      "scripts/run_huginn_mixed_precision_matrix.py": "f64249ff45a0fc0a1813bbc0960bd02b1f98c87e7571dfd697d79822f1b91286",
+      "scripts/run_huginn_presinq_native_screen.py": "592d3a35f85274ff3ef15386dc348f300185409c27c5673c81d54b525a9998d2",
+      "scripts/run_huginn_prespecified_full_queue.py": "266936101d144add1e56290205e01c4cd44e82441e303334402a388b1ef11a07",
+      "scripts/run_huginn_quant_suite.py": "77b3d88fa71b5ccb3710f844da84b2c0e1a87f572dff09ca90c2e311cfeff0a5",
+      "scripts/run_huginn_residual_kv_matrix.py": "3fc23607e30ca800eec5b4040d715a2746f3f3011e627d5cab9329315185e387",
+      "scripts/run_huginn_selective_w4_screen.py": "74f7c15da787e1b7a4c001c5ecac9ffdb6f5f1c9fb62da6d4c1cef1cf5197c7f",
+      "scripts/run_huginn_selective_w4a_matrix.py": "f1ee23f016b8704caaba0d09502e1a6cad237f5b99c912ca2c44242d87d7fd86",
+      "scripts/run_huginn_table5_queue.py": "cbd005233a00c9dc56b8b866845236a47867cc505e2009f47d8d89e914996a7a",
+      "scripts/run_huginn_wa_matrix.py": "1fcc70489a2cdc0a6db05c515b47699e0d4fe49939ab5f0fd8999df41f264502",
+      "scripts/run_int2_accuracy_full.py": "ace6af465d793dd1f33a922fc0de8ba8ac832c519f3f61dc91992b079671be93",
+      "scripts/run_int2_accuracy_smokes.py": "47c987f932ead87cd8715fed7df8c09f6bc18c9385c5d31d5f654c1023e5f87e",
+      "scripts/run_int_kv_table14_15_matrix.py": "b57b985ada939389206cb0d8137f1701216d609a3d231bfb1892c2b95b7434e7",
+      "scripts/run_kv_calibration_robustness.py": "4af1ca5de75b41ac1dc9b9c8967d13993434de4048080cff192859edbba49471",
+      "scripts/run_kv_granularity_matrix.py": "39f394a99e14d43277ba060daf70357038ce42d7075304b1f09149b68158c3f2",
+      "scripts/run_kv_layer_sensitivity.py": "619a4a422338dd2bf7bf892f1dbab6550e79ac9abb16754c9c27fc19ba6e75ea",
+      "scripts/run_layer_channel_delta_ce.py": "cd28a506b86db4d5c717f934e16328d37932c7a0b785b4c6e8613a6453ee73ef",
+      "scripts/run_lm_eval_timed.py": "acdcfdf1b9ffd040e0fadaa45b1942e3d1b548433f135fbe1c84e2fcad0cb6fa",
+      "scripts/run_loop1_remaining_resumed.py": "56b1f88d04e774e4a383097b11640ab3f91bf214fc7b7daaed0286e789ddcc95",
+      "scripts/run_mixed_precision_finalists.py": "0da37cf7c46b974cc1e0835c6ddf8b7dac31b0e50a00b3afec11501e6379e190",
+      "scripts/run_mixed_precision_kv_matrix.py": "4c3dba3d879add89b8b726dfc667d51e2432f78f8f3f283e5b467aa930ec1043",
+      "scripts/run_momentum_ablation_matrix.py": "8d6469daef62830027c2c937e62ac396cdb186af05f330eea4a5a907ae1c6fd8",
+      "scripts/run_native_nvfp4_method_screen.py": "3f8714578306432462d60e6b73b04905bb68433dd6fa7e4ed916dcb56103c28f",
+      "scripts/run_native_wa_bf16_screen.py": "fb102e51d601a442b61afd582e321e5949eac2cfaa0c07fb7ddda53c10e44cc3",
+      "scripts/run_native_wa_export_queue.py": "6abac94be9ddb192cdd306dbf13398502b6f88ee14047027cc71a8d0befa2de3",
+      "scripts/run_ouro_chrome_trace_queue.py": "549b3ddf9a776b2235d704bcf097e725e9c083860f5bd97748937222a8150fa8",
+      "scripts/run_ouro_remaining_queue.py": "d0e698329cc1e986941a062bab9c49b33d43e6fc069b94036c41918ad42d82fb",
+      "scripts/run_ouro_repeatability_queue.py": "3c44894a3b9ef6feb7d91c24d9360dc214ef1bcf320ccd93717a0b27247de05b",
+      "scripts/run_ouro_rotation_math500.py": "3e36126efec9527380e822aaa25cda4caaa1f98595c67abaa3c759892a34e7fb",
+      "scripts/run_ouro_selected_kv.py": "2186aa2dd115633af04fa6b32f996d0edbf4eca8dfe98dd9f60b0219e0b3f159",
+      "scripts/run_ouro_selective_w4_matrix.py": "59de40bab4307007ea068d1ac9a1390c8fb08156bdd6217f19a286cabd4d0b76",
+      "scripts/run_ouro_selective_w4a_matrix.py": "cd59845df8008f51137ee40093b152375afa138e949ff3ced49c752090b3dad4",
+      "scripts/run_ouro_table5_queue.py": "f102ac7637e4652e10da8e7ddc6c5e68228488fefa024841778bdee184fbfaf7",
+      "scripts/run_ouro_vllm_gsm8k.py": "8bdce3830b7e84376b343fda3bec6be9a1efa60a38527e9607d39c2acb30edb6",
+      "scripts/run_packed_affine_sweep.py": "eb8fd503a5aaf83ed3c236aa0ba580fd67e152ec2689e0d868e9f02a54db9b4a",
+      "scripts/run_presinq_native_screen.py": "77d9e4f4ab52e6c155b87afc20e7c674e145b0637ef7053fac0f42dc54123168",
+      "scripts/run_q08_offset_sweep.py": "595e1b0a8a1752a88f7ab3d5a1f9e2ede337768248ae93aa9ec9d6f6d0082d44",
+      "scripts/run_q08_uniform_bits.py": "990ee1bc1c1b5fc5f56d8b787ace89748eaf24645868d6c014a66e47d0b9b259",
+      "scripts/run_residual_activation_matrix.py": "02a21358423ea5ccb2ec83fc3b047073f641494bb61bb6b62837e648ee0693cf",
+      "scripts/run_residual_kv_matrix.py": "392e763fc90e4ea7e3b664e8f70d009a441c0301e87ac52ae5ec033d79013bfc",
+      "scripts/run_selected_wa_kv_validation.py": "3373b6d950c61a7ae8b5af144e01852cfc519e616e4a77416fa955d1f4e3c3f7",
+      "scripts/run_sinq_matrix.py": "3b87d45fb75b69c59a18a9bcb6efde414089671e1c4d579902719cac86a4a6c5",
+      "scripts/run_table10_calibration_queue.py": "e527825b7356cbf5a5d02e0581a8165b38fee0e93874937b48476bbe1cc8179f",
+      "scripts/run_top3_combined_kv.py": "bebfd392c92c744491502ff7fc59b74bbb51dd4be5cf74cc5146c1b4efb5de56",
+      "scripts/select_heatmap_delta_ce_cells.py": "a16a13f8c20b352759769389fb1690cc45b3243cc49824121ee35e313f4cf794",
+      "scripts/select_kv_codec_full_policies.py": "84c5655c547b3faf6052e998c6791ea41105ecfdf5d7137a53757259a75c1991",
+      "scripts/smoke_ouro_nvfp4_vllm.py": "4d6f608d2b9027a6b5f9071511fdef2aaca37e6cad917ec3862c5cf6cc1b3fa6",
+      "scripts/smoke_ouro_recurrent_vllm.py": "bb159fa91c75084a1ff0ae1fdaf6fd65f29a01e6b49d79a16e06c9aec98df749",
+      "scripts/summarize_channel_profile_costs.py": "1835ebab6097d373c9fcab3ccfc0208c2eb392a6c425d1093bb6f5b704b70998",
+      "scripts/summarize_dynamic_lse.py": "66f32fba2b6031ddc80224194a509ec9eadbaebc06ee7832e06c70acfbb1c6e8",
+      "scripts/summarize_graph_experiments.py": "0877b04fb282eb7afafdfcd7e6453b92e6ffbb3521544e7c0719bd87c748f275",
+      "scripts/summarize_int_scale_matrix.py": "4c937809ee46b2576e1aed96996a5525fd2798bcb9b8f83e65eef3dcf0bccb74",
+      "scripts/summarize_kv_cost_analysis.py": "cd409cfad8ecebf0d8d8a3aac1c829342da463f0a36a3ae7ccf63774fef0c563",
+      "scripts/summarize_ouro_attention_precision.py": "a0bf777fc366cc198629c2c467d0fd1f98335bbcca6ae876bb61fe9c68566319",
+      "scripts/summarize_ouro_chrome_traces.py": "47d7beb592fd8049fb61c3e82a0298b4f65f5c35890342f2cc97f73a56843632",
+      "scripts/summarize_ouro_fair_comparison.py": "d16891bdbbb74901cebda757b426e524e447c20bee7c87b36b8ab00bae5f791a",
+      "scripts/summarize_ouro_final_optimization.py": "5222fc60b20210f6786a66005dd32ff3f6a750e5bc92f67654ec325bf0d8f7c7",
+      "scripts/summarize_quantization_goal.py": "73dd960173630d82664d1cd240f71cb329d15fa9842674cd9ca1df115e1370c6",
+      "scripts/summarize_recurrent_full.py": "519bff759541a33cceb3c3513a9537d9272b9b0e8b9c58e3b9f6267bf8b0979a",
+      "scripts/summarize_recurrent_mixed.py": "406b963dbb92587f781402a6b386bdf55d98bf053aab4c0f25432c931510662c",
+      "scripts/summarize_recurrent_screen.py": "68fcc410a93a0d667b97e7ab2a334c55eaa1d7250a8d3ad8f245ce8988db7f9d",
+      "scripts/summarize_recurrent_xqa.py": "c4b3718a49fd016a7d1b7ef3d1772cd0f31ef038f5a596ea0eb63460e035bc62",
+      "scripts/summarize_runpod_benchmarks.py": "9d18c35c572aed7b4ca1ba8f08bd79db2094bcee61252733f7c3c1a19189cf5f",
+      "scripts/summarize_scan_full.py": "7c10989917660da80a64ae4414dd12bec35d9d8e3d24a0b0d6670081bc3fc234",
+      "scripts/summarize_xqa_exploration.py": "9abeef642993d443c99a5a98970c89688326c45a7245c0cad617442506664f4a",
+      "scripts/trace_channel_capacity_schedule.py": "6fdb5972552f69c96318764f1e50fda3d08487db3b471c7d56f8ebaafec905ba",
+      "scripts/validate_native_kv.py": "cb726f0a16755d968317cb133bf232de1b0135dfbc0fa2f73efab2557af5775f",
+      "scripts/validate_nvfp4_modelopt_hopper.py": "079cf32815542feb97b0784696bbac1488c57d562c70905cb67355e74160e8f8",
+      "scripts/validate_pre_sinq_checkpoint.py": "5f6a7d4daed934b88e30eb4dde79f62a272782b69a35a220f4899be966af8a2d",
+      "scripts/validate_reqat_e1m2_modelopt.py": "31a076de14ab0bd99facfa7a2ca19eac2aa947c2872fe64c70f08e87f40b0e0c",
+      "plugins/ouro_quant_vllm/pyproject.toml": "c16833fa241d6219d645f79979c5df2bc4a8c3494a74a35d4c8ac38d59e96fcc",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/__init__.py": "09961d7d247f2e2667eaa45747270f53172d32422e3a09391a3ad65d149822f8",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/channel_requests.py": "3389cf929bedb1dd3d4d38a7dec8ab598cba3fa05fef8746f3d32d9fa34c2474",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/fused_norm.py": "ea5b02eda1eed9280729ac2801ecefb8fc41fd82002bd76bf9ecce5c7c586e54",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/fused_rope_cache.py": "7765006ecd10e9b8562f5d479cff12f4471c807bab271954a699bb381dc97c83",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/kv_rotation.py": "b7d4ff4342d42a07d7791180c6ffad931a8d283ab0b46c7b1e4ae32865c0d436",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/loopq_bridge.py": "be35be25a58efdc2fb0ebc7f7101b01d9c8b0aebd66b0f0a60c21fe5557cf468",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/model.py": "806182c39a5693efd7895a607be17c44ccbc2b74d4e6624d90596c7a4cfb1f0e",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/native_attention.py": "f2ddc785386102a38a40cc8b108ab2c4bfb7fe99cd3dcbe2b400f747b59483aa",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/nvfp4_qdq.py": "b55c0270fdb6131848456141261d1c6b984f3f88ed31ac561ee7f981cb6a593c",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_affine_attention.py": "0a9a0a1018341df5261254fdc9fabf1e1534381d91ab7b91a8a7c8ef6e0bccc2",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_rotated_attention.py": "d8f7b9ec10f0c1c15e6043fc26444b4a6a376bfffdbcd8232e0ce577e270a01c",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_rotated_table1_attention.py": "7ae22f08c2efd240fb4c549943eaaae972a3161f482b91cbb5194bda3fd6b1a7",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/recurrent_attention.py": "2e6ff15630c0757a9a36ca14818535f7a605be2800de8083841493ba25524dc3",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/residual_activation.py": "8cca2e8ad00594220b2658c4be851efdc9627506c01900e8647c89463b8bc877",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/residual_kv.py": "4cad104ba295e294f87eb27eada1d968be46a7d5b4b0c08edb52e8bbebfc8e49",
+      "plugins/huginn_vllm_validated/pyproject.toml": "4538f596380c7d8487189accd14597cb2ed237dd76bb719a7057a95abefb4f14",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/__init__.py": "5e18f53c2fdb7911b6cc921decc552d1fa92369fc115e1a1199288e58bcb16cc",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/attention_bank.py": "8482b5175982303d7d999a5e0fdf71e2e6f09d5f34a88e57a6bdeb3df58a7dc8",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/grouped_attention.py": "1e7df3c0c3f7cebc40cb977c11926a0b2c032653c20454a160a7ecbc21fd7b63",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/loopq_bridge.py": "5cb3b60e24a2c683976d6229dd89135a04e3167ede696742a55a5dd327d49f7e",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/model.py": "252b5b87d06ee17896fc91945a645c1e528290c70be7e2129bf9f2271177e383",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/native_attention.py": "93843837ccd9b3393a080391d9c013a9586304af1acc1d850e52e8fbec7d2294",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/runtime.py": "c9b71f1f48d75475ea299eab01d5bcf6f38445a7c7eec62fe2d5205b47d6c218",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/wavefront.py": "a74f9b084a9e2708c1211dc9fb6cb20de23d61e17868e0967fcf0e6cd27eb125"
+    },
+    "scope": "repository source at launch; installed dependencies are identified separately by environment metadata",
+    "changed_during_run": [],
+    "source_tree_unchanged_at_exit": true
+  },
+  "result_path": "/workspace/loopq_data/curve_a4_s0/eval_final/arc_challenge/harness/ByteDance__Ouro-1.4B/results_2026-09-17T00-10-49.223641.json",
+  "samples": {
+    "arc_challenge": {
+      "original": 1172,
+      "effective": 1172
+    }
+  },
+  "metrics": {
+    "arc_challenge": {
+      "name": "arc_challenge",
+      "alias": "arc_challenge",
+      "sample_len": 1172,
+      "acc,none": 0.47696245733788395,
+      "acc_stderr,none": 0.014595873205358269,
+      "acc_norm,none": 0.4906143344709898,
+      "acc_norm_stderr,none": 0.014608816322065
+    }
+  },
+  "task_configs": {
+    "arc_challenge": {
+      "task": "arc_challenge",
+      "dataset_path": "allenai/ai2_arc",
+      "dataset_name": "ARC-Challenge",
+      "training_split": "train",
+      "validation_split": "validation",
+      "test_split": "test",
+      "doc_to_text": "Question: {{question}}\nAnswer:",
+      "doc_to_target": "{{choices.label.index(answerKey)}}",
+      "unsafe_code": false,
+      "doc_to_choice": "{{choices.text}}",
+      "description": "",
+      "target_delimiter": " ",
+      "fewshot_delimiter": "\n\n",
+      "fewshot_config": {
+        "sampler": "default",
+        "split": null,
+        "process_docs": null,
+        "fewshot_indices": null,
+        "samples": null,
+        "doc_to_text": "Question: {{question}}\nAnswer:",
+        "doc_to_choice": "{{choices.text}}",
+        "doc_to_target": "{{choices.label.index(answerKey)}}",
+        "gen_prefix": null,
+        "fewshot_delimiter": "\n\n",
+        "target_delimiter": " "
+      },
+      "num_fewshot": 0,
+      "metric_list": [
+        {
+          "metric": "acc",
+          "aggregation": "mean",
+          "higher_is_better": true
+        },
+        {
+          "metric": "acc_norm",
+          "aggregation": "mean",
+          "higher_is_better": true
+        }
+      ],
+      "output_type": "multiple_choice",
+      "repeats": 1,
+      "should_decontaminate": true,
+      "doc_to_decontamination_query": "Question: {{question}}\nAnswer:",
+      "metadata": {
+        "version": 1.0,
+        "pretrained": "ByteDance/Ouro-1.4B",
+        "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+        "trust_remote_code": true,
+        "dtype": "bfloat16",
+        "tensor_parallel_size": 1,
+        "max_model_len": 4096,
+        "gpu_memory_utilization": 0.9,
+        "enforce_eager": true,
+        "max_num_seqs": 32,
+        "max_num_batched_tokens": 4096,
+        "enable_prefix_caching": false,
+        "seed": 0,
+        "config_source": "/root/venv-LT/lib/python3.11/site-packages/lm_eval/tasks/arc/arc_challenge.yaml"
+      }
+    }
+  }
+}

loopq_quantization/results/loopq_w4a4_completed/arc_challenge/timing.json

@@ -0,0 +1,9 @@
+{
+  "model_init_seconds": 34.56173160113394,
+  "generate_until_seconds": 0.0,
+  "generated_samples": 0,
+  "generated_tokens": 0,
+  "generation_calls": 0,
+  "completed": true,
+  "process_seconds": 138.37667733803391
+}

loopq_quantization/results/loopq_w4a4_completed/hellaswag/harness/ByteDance__Ouro-1.4B/results_2026-09-17T00-06-20.554115.json

@@ -0,0 +1,153 @@
+{
+  "results": {
+    "hellaswag": {
+      "name": "hellaswag",
+      "alias": "hellaswag",
+      "sample_len": 10042,
+      "acc,none": 0.5041824337781319,
+      "acc_stderr,none": 0.00498960683837106,
+      "acc_norm,none": 0.6809400517825135,
+      "acc_norm_stderr,none": 0.004651597209993087
+    }
+  },
+  "group_subtasks": {},
+  "configs": {
+    "hellaswag": {
+      "task": "hellaswag",
+      "dataset_path": "Rowan/hellaswag",
+      "training_split": "train",
+      "validation_split": "validation",
+      "process_docs": "def process_docs(dataset: datasets.Dataset) -> datasets.Dataset:\n    def _process_doc(doc):\n        ctx = doc[\"ctx_a\"] + \" \" + doc[\"ctx_b\"].capitalize()\n        out_doc = {\n            \"query\": preprocess(doc[\"activity_label\"] + \": \" + ctx),\n            \"choices\": [preprocess(ending) for ending in doc[\"endings\"]],\n            \"gold\": int(doc[\"label\"]),\n        }\n        return out_doc\n\n    return dataset.map(_process_doc)\n",
+      "doc_to_text": "{{query}}",
+      "doc_to_target": "{{label}}",
+      "unsafe_code": false,
+      "doc_to_choice": "choices",
+      "description": "",
+      "target_delimiter": " ",
+      "fewshot_delimiter": "\n\n",
+      "fewshot_config": {
+        "sampler": "default",
+        "split": null,
+        "process_docs": "<function process_docs at 0x783c92691e40>",
+        "fewshot_indices": null,
+        "samples": null,
+        "doc_to_text": "{{query}}",
+        "doc_to_choice": "choices",
+        "doc_to_target": "{{label}}",
+        "gen_prefix": null,
+        "fewshot_delimiter": "\n\n",
+        "target_delimiter": " "
+      },
+      "num_fewshot": 0,
+      "metric_list": [
+        {
+          "metric": "acc",
+          "aggregation": "mean",
+          "higher_is_better": true
+        },
+        {
+          "metric": "acc_norm",
+          "aggregation": "mean",
+          "higher_is_better": true
+        }
+      ],
+      "output_type": "multiple_choice",
+      "repeats": 1,
+      "should_decontaminate": false,
+      "metadata": {
+        "version": 1.0,
+        "pretrained": "ByteDance/Ouro-1.4B",
+        "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+        "trust_remote_code": true,
+        "dtype": "bfloat16",
+        "tensor_parallel_size": 1,
+        "max_model_len": 4096,
+        "gpu_memory_utilization": 0.9,
+        "enforce_eager": true,
+        "max_num_seqs": 32,
+        "max_num_batched_tokens": 4096,
+        "enable_prefix_caching": false,
+        "seed": 0,
+        "config_source": "/root/venv-LT/lib/python3.11/site-packages/lm_eval/tasks/hellaswag/hellaswag.yaml"
+      }
+    }
+  },
+  "versions": {
+    "hellaswag": 1.0
+  },
+  "n-shot": {
+    "hellaswag": 0
+  },
+  "higher_is_better": {
+    "hellaswag": {
+      "acc": true,
+      "acc_norm": true
+    }
+  },
+  "n-samples": {
+    "hellaswag": {
+      "original": 10042,
+      "effective": 10042
+    }
+  },
+  "config": {
+    "model": "vllm",
+    "model_args": {
+      "pretrained": "ByteDance/Ouro-1.4B",
+      "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+      "trust_remote_code": true,
+      "dtype": "bfloat16",
+      "tensor_parallel_size": 1,
+      "max_model_len": 4096,
+      "gpu_memory_utilization": 0.9,
+      "enforce_eager": true,
+      "max_num_seqs": 32,
+      "max_num_batched_tokens": 4096,
+      "enable_prefix_caching": false,
+      "seed": 0
+    },
+    "batch_size": "auto",
+    "batch_sizes": [],
+    "device": "cuda:0",
+    "use_cache": null,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "gen_kwargs": {},
+    "random_seed": 0,
+    "numpy_seed": 0,
+    "torch_seed": 0,
+    "fewshot_seed": 0
+  },
+  "git_hash": null,
+  "date": 1789602217.133009,
+  "pretty_env_info": "PyTorch version: 2.8.0+cu128\nIs debug build: False\nCUDA used to build PyTorch: 12.8\nROCM used to build PyTorch: N/A\n\nOS: Ubuntu 22.04.5 LTS (x86_64)\nGCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0\nClang version: Could not collect\nCMake version: Could not collect\nLibc version: glibc-2.35\n\nPython version: 3.11.14 | packaged by conda-forge | (main, Oct 13 2025, 14:09:32) [GCC 14.3.0] (64-bit runtime)\nPython platform: Linux-6.8.0-124-generic-x86_64-with-glibc2.35\nIs CUDA available: True\nCUDA runtime version: Could not collect\nCUDA_MODULE_LOADING set to: LAZY\nGPU models and configuration: \nGPU 0: NVIDIA A40\nGPU 1: NVIDIA A40\n\nNvidia driver version: 580.159.04\ncuDNN version: Could not collect\nHIP runtime version: N/A\nMIOpen runtime version: N/A\nIs XNNPACK available: True\n\nCPU:\nArchitecture:                            x86_64\nCPU op-mode(s):                          32-bit, 64-bit\nAddress sizes:                           52 bits physical, 57 bits virtual\nByte Order:                              Little Endian\nCPU(s):                                  96\nOn-line CPU(s) list:                     0-95\nVendor ID:                               GenuineIntel\nModel name:                              Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz\nCPU family:                              6\nModel:                                   106\nThread(s) per core:                      2\nCore(s) per socket:                      24\nSocket(s):                               2\nStepping:                                6\nCPU max MHz:                             3500.0000\nCPU min MHz:                             800.0000\nBogoMIPS:                                5600.00\nFlags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect wbnoinvd dtherm ida arat pln pts vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities\nVirtualization:                          VT-x\nL1d cache:                               2.3 MiB (48 instances)\nL1i cache:                               1.5 MiB (48 instances)\nL2 cache:                                60 MiB (48 instances)\nL3 cache:                                72 MiB (2 instances)\nNUMA node(s):                            2\nNUMA node0 CPU(s):                       0-23,48-71\nNUMA node1 CPU(s):                       24-47,72-95\nVulnerability Gather data sampling:      Mitigation; Microcode\nVulnerability Indirect target selection: Mitigation; Aligned branch/return thunks\nVulnerability Itlb multihit:             Not affected\nVulnerability L1tf:                      Not affected\nVulnerability Mds:                       Not affected\nVulnerability Meltdown:                  Not affected\nVulnerability Mmio stale data:           Mitigation; Clear CPU buffers; SMT vulnerable\nVulnerability Reg file data sampling:    Not affected\nVulnerability Retbleed:                  Not affected\nVulnerability Spec rstack overflow:      Not affected\nVulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl\nVulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization\nVulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop\nVulnerability Srbds:                     Not affected\nVulnerability Tsa:                       Not affected\nVulnerability Tsx async abort:           Not affected\nVulnerability Vmscape:                   Not affected\n\nVersions of relevant libraries:\n[pip3] numpy==2.2.6\n[pip3] nvidia-cublas-cu12==12.8.4.1\n[pip3] nvidia-cuda-cupti-cu12==12.8.90\n[pip3] nvidia-cuda-nvrtc-cu12==12.8.93\n[pip3] nvidia-cuda-runtime-cu12==12.8.90\n[pip3] nvidia-cudnn-cu12==9.10.2.21\n[pip3] nvidia-cufft-cu12==11.3.3.83\n[pip3] nvidia-curand-cu12==10.3.9.90\n[pip3] nvidia-cusolver-cu12==11.7.3.90\n[pip3] nvidia-cusparse-cu12==12.5.8.93\n[pip3] nvidia-cusparselt-cu12==0.7.1\n[pip3] nvidia-nccl-cu12==2.27.3\n[pip3] nvidia-nvjitlink-cu12==12.8.93\n[pip3] nvidia-nvtx-cu12==12.8.90\n[pip3] torch==2.8.0\n[pip3] torchaudio==2.8.0\n[pip3] torchvision==0.23.0\n[pip3] triton==3.4.0\n[conda] Could not collect",
+  "transformers_version": "4.56.2",
+  "lm_eval_version": "0.4.13.dev0",
+  "upper_git_hash": null,
+  "tokenizer_pad_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "tokenizer_eos_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "tokenizer_bos_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "eot_token_id": 0,
+  "max_length": 4096,
+  "task_hashes": {
+    "hellaswag": "edcc7edd27a555d3f7cbca0641152b2c5e4eb6eb79c5e62d7fe5887f47814323"
+  },
+  "model_source": "vllm",
+  "model_name": "ByteDance/Ouro-1.4B",
+  "model_name_sanitized": "ByteDance__Ouro-1.4B",
+  "system_instruction": null,
+  "system_instruction_sha": null,
+  "fewshot_as_multiturn": null,
+  "chat_template": null,
+  "chat_template_sha": null,
+  "total_evaluation_time_seconds": "1369.5787835847586"
+}

loopq_quantization/results/loopq_w4a4_completed/hellaswag/harness/ByteDance__Ouro-1.4B/samples_hellaswag_2026-09-17T00-06-20.554115.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b11173df38be2f87ecda9fbab9a49faa708b68aad9490f7f6b45cce9e081253
+size 43664010

loopq_quantization/results/loopq_w4a4_completed/hellaswag/installed_task_snapshot/README.md

@@ -0,0 +1,49 @@
+# HellaSwag
+
+### Paper
+
+Title: `HellaSwag: Can a Machine Really Finish Your Sentence?`
+
+Abstract: https://arxiv.org/abs/1905.07830
+
+Recent work by Zellers et al. (2018) introduced a new task of commonsense natural language inference: given an event description such as "A woman sits at a piano," a machine must select the most likely followup: "She sets her fingers on the keys." With the introduction of BERT, near human-level performance was reached. Does this mean that machines can perform human level commonsense inference?
+In this paper, we show that commonsense inference still proves difficult for even state-of-the-art models, by presenting HellaSwag, a new challenge dataset. Though its questions are trivial for humans (>95% accuracy), state-of-the-art models struggle (<48%). We achieve this via Adversarial Filtering (AF), a data collection paradigm wherein a series of discriminators iteratively select an adversarial set of machine-generated wrong answers. AF proves to be surprisingly robust. The key insight is to scale up the length and complexity of the dataset examples towards a critical 'Goldilocks' zone wherein generated text is ridiculous to humans, yet often misclassified by state-of-the-art models.
+Our construction of HellaSwag, and its resulting difficulty, sheds light on the inner workings of deep pretrained models. More broadly, it suggests a new path forward for NLP research, in which benchmarks co-evolve with the evolving state-of-the-art in an adversarial way, so as to present ever-harder challenges.
+
+Homepage: `https://rowanzellers.com/hellaswag/`
+
+
+### Citation
+
+```
+@inproceedings{zellers2019hellaswag,
+    title={HellaSwag: Can a Machine Really Finish Your Sentence?},
+    author={Zellers, Rowan and Holtzman, Ari and Bisk, Yonatan and Farhadi, Ali and Choi, Yejin},
+    booktitle ={Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},
+    year={2019}
+}
+```
+
+### Groups and Tasks
+
+#### Groups
+
+- Not part of a group yet
+
+#### Tasks
+
+- `hellaswag`
+
+
+### Checklist
+
+For adding novel benchmarks/datasets to the library:
+* [x] Is the task an existing benchmark in the literature?
+  * [x] Have you referenced the original paper that introduced the task?
+  * [x] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
+
+
+If other tasks on this dataset are already supported:
+* [ ] Is the "Main" variant of this task clearly denoted?
+* [ ] Have you provided a short sentence in a README on what each new variant adds / evaluates?
+* [ ] Have you noted which, if any, published evaluation setups are matched by this variant?

loopq_quantization/results/loopq_w4a4_completed/hellaswag/installed_task_snapshot/hellaswag.yaml

@@ -0,0 +1,22 @@
+tag:
+  - multiple_choice
+task: hellaswag
+dataset_path: Rowan/hellaswag
+dataset_name: null
+output_type: multiple_choice
+training_split: train
+validation_split: validation
+test_split: null
+process_docs: !function utils.process_docs
+doc_to_text: "{{query}}"
+doc_to_target: "{{label}}"
+doc_to_choice: "choices"
+metric_list:
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true
+  - metric: acc_norm
+    aggregation: mean
+    higher_is_better: true
+metadata:
+  version: 1.0

loopq_quantization/results/loopq_w4a4_completed/hellaswag/installed_task_snapshot/utils.py

@@ -0,0 +1,25 @@
+import re
+
+import datasets
+
+
+def preprocess(text):
+    text = text.strip()
+    # NOTE: Brackets are artifacts of the WikiHow dataset portion of HellaSwag.
+    text = text.replace(" [title]", ". ")
+    text = re.sub("\\[.*?\\]", "", text)
+    text = text.replace("  ", " ")
+    return text
+
+
+def process_docs(dataset: datasets.Dataset) -> datasets.Dataset:
+    def _process_doc(doc):
+        ctx = doc["ctx_a"] + " " + doc["ctx_b"].capitalize()
+        out_doc = {
+            "query": preprocess(doc["activity_label"] + ": " + ctx),
+            "choices": [preprocess(ending) for ending in doc["endings"]],
+            "gold": int(doc["label"]),
+        }
+        return out_doc
+
+    return dataset.map(_process_doc)

loopq_quantization/results/loopq_w4a4_completed/hellaswag/run_metadata.json

@@ -0,0 +1,722 @@
+{
+  "status": "completed",
+  "task": "hellaswag",
+  "configuration": "loopq_w4a4",
+  "contract": {
+    "mode": "loopq",
+    "activation_bits": 4,
+    "artifact": "/workspace/loopq_data/curve_a4_s0/eval_final/artifact.pt",
+    "artifact_sha256": "7cb498143aeb807e5cb896f195066fa07e38a353406a10de36ffaf46b4fcabdb",
+    "ablation": null,
+    "parity_evidence": {
+      "path": "/workspace/loopq_data/curve_a4_s0/eval_final/parity_screen.json",
+      "sha256": "0492028e64eaf16242d69d32bec9a7783e3637875ddaae3b986889c86a1b04ed",
+      "acceptance": {
+        "passed": false,
+        "reasons": [
+          "four-loop hidden-state parity failed",
+          "logprob error exceeds tolerance",
+          "top1 or top-1000 comparison incomplete"
+        ],
+        "max_logprob_error": 1.25,
+        "max_hidden_error": 0.06,
+        "tie_threshold": 0.25,
+        "rows_excused_as_reference_tie": 0,
+        "scope": "six_or_more_single_sequence_prefills; decode/batch/long_context still require A40 validation",
+        "advisory": true
+      }
+    }
+  },
+  "model_args": {
+    "pretrained": "ByteDance/Ouro-1.4B",
+    "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+    "trust_remote_code": true,
+    "dtype": "bfloat16",
+    "tensor_parallel_size": 1,
+    "max_model_len": 4096,
+    "gpu_memory_utilization": 0.9,
+    "enforce_eager": true,
+    "max_num_seqs": 32,
+    "max_num_batched_tokens": 4096,
+    "enable_prefix_caching": false,
+    "seed": 0
+  },
+  "fewshot": 0,
+  "seed": 0,
+  "smoke_limit": null,
+  "study_evaluation": false,
+  "protocol_note": "LoopQ W4A4, calibration run to completion (1500 final optimization steps) with singular values held in log space so they cannot reach zero. The paper does not specify how the Kronecker transforms are kept invertible. Deployment prefill-parity gate is unreachable at W4A4 by any implementation and is attached as advisory. Explicit local protocol: zero-shot except MMLU 5-shot; plain likelihood scoring, no chat template; installed pinned harness prompts.",
+  "metric_note": "acc_norm where available, otherwise acc; retain all metrics",
+  "exact_paper_protocol_reproduction": false,
+  "disclosure": "Single explicit-seed run; paper reports ten-run accuracy averages. Dataset/harness choices must be compared before claiming reproduction.",
+  "command": [
+    "/root/venv-LT/bin/python",
+    "/workspace/LT_fix/scripts/run_lm_eval_timed.py",
+    "--model",
+    "vllm",
+    "--model_args",
+    "{\"pretrained\": \"ByteDance/Ouro-1.4B\", \"revision\": \"e3b1e0993b1231a51d6069a870476dda4162c00a\", \"trust_remote_code\": true, \"dtype\": \"bfloat16\", \"tensor_parallel_size\": 1, \"max_model_len\": 4096, \"gpu_memory_utilization\": 0.9, \"enforce_eager\": true, \"max_num_seqs\": 32, \"max_num_batched_tokens\": 4096, \"enable_prefix_caching\": false, \"seed\": 0}",
+    "--tasks",
+    "hellaswag",
+    "--num_fewshot",
+    "0",
+    "--seed",
+    "0,0,0,0",
+    "--batch_size",
+    "auto",
+    "--output_path",
+    "/workspace/loopq_data/curve_a4_s0/eval_final/hellaswag/harness",
+    "--log_samples"
+  ],
+  "environment": {
+    "torch": "2.8.0",
+    "vllm": "0.10.2",
+    "transformers": "4.56.2",
+    "lm_eval": "0.4.13.dev0",
+    "datasets": "5.0.1"
+  },
+  "returncode": 0,
+  "wall_clock_seconds": 1382.181919325143,
+  "source_provenance": {
+    "snapshot": "/workspace/loopq_data/curve_a4_s0/eval_final/hellaswag/source_snapshot",
+    "sha256_at_launch": {
+      "loopQ/adapters/__init__.py": "7f28e14f7778bafb750404e0e54fb05fe17b61997e06584ca55ee44da3694ba7",
+      "loopQ/adapters/huginn.py": "fc22beafc461a7a0f85e6a7e3291a60749d5a11ea56cceabc448239dd3be9afe",
+      "loopQ/adapters/ouro.py": "e571df0cbbe853de8d1c593f9ed6c032e9099008363ee57e1f280b770ae49c14",
+      "loopQ/configs/flatquant_reference.yaml": "35f507d7389c3f04a7c1f86f7ae367630c846ec730661d420639f5f19b0c4074",
+      "loopQ/configs/huginn_mapping.yaml": "5db252e7cacb42169e94492e02509a3114ebca4747b32590110ba58cbbaea124",
+      "loopQ/configs/loopq_compatibility_decisions.yaml": "3d1b6d508db78b53f2c6e405da3b8ea6a7893303bd81bc2a90e0b3567367fd35",
+      "loopQ/configs/ouro_calibration.yaml": "fc832996fcdda085290565b657c6fa1824dc0b1646a66d109fe38b22f0a8eac5",
+      "loopQ/configs/ouro_mapping.yaml": "87ca4718e200539412f366fbad52a1a46932ad2cc26e0f08585830d04f50b9ad",
+      "loopQ/configs/paper_ouro_reference_metrics.json": "ecc4b3e028b1191a458b446582895d4f6871922a397f15375a34e396bf70ad89",
+      "loopQ/configs/paper_specification.yaml": "c812bb40398d9183f522c07e32dabf3fcfdd43e4d52790b9af296fb11d5abed1",
+      "loopQ/configs/parity_prompts.json": "b56e2e4048083459997c2864e3e6320fc68e1625676ed09f207a81bafee9b2e0",
+      "loopQ/configs/reproduction.json": "5204e346d98f96e46b0f94814e2f291198887ace85a7c827932f002560c350c3",
+      "loopQ/loopq/__init__.py": "00cf79e9673051148513cfc25c52c59ef72118d288e2ec640a77eaf8a69d7213",
+      "loopQ/loopq/accounting.py": "ab24ef8b10a0bc85e03827e882f92e1d7bb3e33fbd146f7f9561651842a03fb0",
+      "loopQ/loopq/calibrate.py": "5018098ddf0c44a682efef85b9f82ab2e4f4f911efa48982083d6c3a0db98594",
+      "loopQ/loopq/cta.py": "e1cf8cfec967157a90d9f54e956b221b67aef9ed742968605d2d4abb1f453b8d",
+      "loopQ/loopq/evaluation_provenance.py": "87e0d6f329dae65759893063cf508682902ae4cf5772aec28ed1692a08da02cf",
+      "loopQ/loopq/experiment_contract.py": "a0f4487092a72eb5fe808880d09a6fafc9b86e5fc4a2e2b3b9e7d5177d27a620",
+      "loopQ/loopq/experiments.py": "252644613f076890e5f2b849c4f95570e33a9cfa69569df90b082c26788f5a5b",
+      "loopQ/loopq/export.py": "3ca272d220bbc93d2e7c70b06df2cef93184e48bc695c00b97e834991469ef57",
+      "loopQ/loopq/gpu_admission.py": "feb940b5cc2355f1881d031f9b22226d7692f70f792c1b1233aeec5570535c76",
+      "loopQ/loopq/huginn_calibration.py": "7a95635b9d9777c9b0f04f51566f9ed668154ae9be231e19f3b7ac18acef98ca",
+      "loopQ/loopq/journal.py": "f9bf8342a004959bbdb25b3557577e5271e7335f53107d7a007ad512fa55bbcf",
+      "loopQ/loopq/las.py": "a179c62e57fccafc21dabb13a68832f8bc8c9d6b0a8da1a2c2348bfb09c47a3c",
+      "loopQ/loopq/numerical_contract.py": "913685841222a9f7ff503ca9b8962cd976f39bda38e2a20f7d3e6b0c09798bc5",
+      "loopQ/loopq/objective.py": "d60f1d6ce18e004f89cb91f680ed6dd4b28083dda8125745a1bd694389c40e50",
+      "loopQ/loopq/ouro_calibration.py": "db88d48b9646c421e4fd354e8de61a7784bdedcbac0b47e1c8de01aa8c2395de",
+      "loopQ/loopq/packed_bundle.py": "5fe29dee51fe28849696de298d3bc8f7036b78c44ec2617747ada370e135287f",
+      "loopQ/loopq/packed_weight.py": "8bbd5a3dc0c8992e4dd30ce962deb0e0745604b1e189a508e1f70aee1e1acdbc",
+      "loopQ/loopq/packed_worker.py": "239febb2ed17d9d88cb068d2edac87f13c2f5c579995a842cdfb7ab6b1e536aa",
+      "loopQ/loopq/parity_worker.py": "b6854262f3b44e4970214aa3e8ccc5a8ad1fa032a9ae5314a15514e132540502",
+      "loopQ/loopq/paths.py": "5a9478359f3a4b252d5af599c9c7f68650d10b995f2e9314210ac6e670a364d0",
+      "loopQ/loopq/provenance.py": "387f82f3efe26ea478476d7087db441a2552f345de35ce2afc5723eb4d21abd6",
+      "loopQ/loopq/quantization.py": "78a3016ad8b2d53ae807bb18b71da8e7d8ce6df8d0881331a80776a247c1142e",
+      "loopQ/loopq/readiness.py": "837a59bc6ac641ea2140876fd3ae34afc30173b91791852ba55dafd4bdd9cdd8",
+      "loopQ/loopq/saved_tensors.py": "ee158cc09d8116c09922b940d3219d7d22d3f271b9c11e8b623b393cecb2af03",
+      "loopQ/loopq/sharing_gap.py": "aa6e1bdb0f365fdde0e9ed4674a08b0f2c0df0a0693dd28dc5c094f8f86118a5",
+      "loopQ/loopq/trajectory_diagnostics.py": "de78ef945d1fdcd036460ad5947864f73a5d63a5f88a1a6c86691821b2dddcac",
+      "loopQ/loopq/transforms.py": "d754f21df55fde2548872d25afa9a4fe75082fdc937a3ba65b474a0673be12a6",
+      "loopQ/scripts/benchmark_calibration_memory.py": "afdaf39cde1e839924baf1cfd0a4d83df5ee0af40d6e7659cb6f3645e3ebb408",
+      "loopQ/scripts/calibrate_huginn.py": "ba44cc894b5a9987cb2f3a1382053d907f205a2d7608b44b7714fc410c142756",
+      "loopQ/scripts/calibrate_ouro.py": "5c81cb8ca6a92a6da4fe1b7e5d7615830e359d93a4eedf013fbb32d5f624ec9e",
+      "loopQ/scripts/diagnose_ouro_quality.py": "962f5ac918bea5cb029c78f41b96b886013e8233d5bf1d67ec87d8b488b9977c",
+      "loopQ/scripts/export_huginn_artifact.py": "20dbf608320f85daddbc843b0ac1196234c20822e2418f42b66327ddcd8186f7",
+      "loopQ/scripts/export_ouro_artifact.py": "1fe2edfa0fc09cf896a1fab149067160829744d3f7fd03d2c521b6ba38d19a12",
+      "loopQ/scripts/export_packed_ouro_weights.py": "a49b8646639682012340f908ec86efbe12a369343c984963c2314991c31a4ead",
+      "loopQ/scripts/freeze_lambada_inputs.py": "b15138ae82a52461f31cbb669d18f878fd98f9eab93ac7db3efa87f594afb97b",
+      "loopQ/scripts/inspect_calibration.py": "6514bb2fb9fda6ef6b5335035dae910cb4bf7ec36dde5a844c7cf38e0fd07f7a",
+      "loopQ/scripts/measure_ouro_trajectory.py": "06bf7ea56bb6626b5947e8f43d2b7ddc60ad942d3997ac1c2b7795db4672c081",
+      "loopQ/scripts/report_accounting.py": "3564bf5cb9e4f8133257cb415ec679f007090f0c7afe85fe1bca4b469b5ba5e6",
+      "loopQ/scripts/reproduction.py": "6a28f7a4cb1f4a9f4171fcf466b36b103761bc38bde63e3afefe8a25aca86fb1",
+      "loopQ/scripts/run_gsm8k.py": "11b220ad74277b5da6becbf18eaaeb4725d6ecb07330b1b15cd4dc4c29eeefdb",
+      "loopQ/scripts/run_paper_benchmark.py": "021ee1451abff633bc0b77f04c13e1cd9dae7bebfc86ebe962890015a74a6bf8",
+      "loopQ/scripts/smoke_huginn_vllm.py": "9311b038ce325a3f050a852addb6a9c550a09ace19f29ee3370439be9047e29b",
+      "loopQ/scripts/smoke_ouro_vllm.py": "39311a193900e7d1db491e29fcdd83ee88ad1bdda602025580f61bd5ae757203",
+      "loopQ/scripts/verify_runtime_parity.py": "1e0e1648fe5673abaca26d98efffac99cc3b2ede7ecb59423df04147e537e4e0",
+      "loopQ/tests/conftest.py": "56cfc62ddfe4ccab80288f60c8f3948b6d8b8d8f1d2e37c50f9c9577c4679235",
+      "loopQ/tests/test_alignment_contracts.py": "615d2e77c187411d55892fa875835ea280a9b6ab775835ed30459482c298f066",
+      "loopQ/tests/test_calibrate.py": "d9bddf96ebcd0c5bdd0dd866f17e045d672abd73360e5e0771775edc4e67c8c8",
+      "loopQ/tests/test_cta.py": "afe35581d60dcd0b5ff1566ffe3a8dc332003cd031d9c9e788cff1c2f62f2307",
+      "loopQ/tests/test_evaluation_provenance.py": "62765331e2e5b0e400c51b0d7caba5cc70f3060ac60fdfa53e89fc0eea293933",
+      "loopQ/tests/test_experiment_setup.py": "840680d43db300613ccb55409ce9bd142f2eb82176d697813194648de2c57484",
+      "loopQ/tests/test_export.py": "7bf542a1e606c7be517c2485dcad3b8e8ec773f53e7374656564ce33a78a6b8a",
+      "loopQ/tests/test_gpu_admission.py": "5f0154e0e7d80e1de684b51219b3ee1b52d81481d062de13a563eab1db17fce6",
+      "loopQ/tests/test_gsm8k_driver.py": "054d0284905dc8479ceb6c28e2a418d94a692762d45573902a8bb0278abd2821",
+      "loopQ/tests/test_huginn_adapter.py": "23f7b5d2fea8c9526553e979a5c44e466fe74d664cf6f0a05b8c66434dc48940",
+      "loopQ/tests/test_huginn_calibration.py": "bbdc7f0e10f23042f6d52acf14456d039b030d945fb4d3a8c72d9d6063242a93",
+      "loopQ/tests/test_huginn_export.py": "896eee3b771ee2a2b0dbdbd9aa07eb9918718c7dc9e69e0169b3d4ab22e50229",
+      "loopQ/tests/test_huginn_vllm_bridge.py": "c483f52999ca955b4c8b17de252e957bf7a878bfa7e5fa18330c7fed3f249c9b",
+      "loopQ/tests/test_journal.py": "b3331cbff1add620c8e1d098578a9d79f81e2c1d28a2e0c0b4a7a7a0c7ba1589",
+      "loopQ/tests/test_las.py": "88139ae1f23a53d07eb3497b88fe3f22980d6b8c001232fc3b629a57663935e4",
+      "loopQ/tests/test_objective.py": "fb206b4844e1f911e6665ccd5d99e643171810a30da50a80be64d227019e9b88",
+      "loopQ/tests/test_ouro_adapter.py": "9a52daa93ea65a75fc9edfd154ab27058c6a2567d113cab25e6d5a9e6592d986",
+      "loopQ/tests/test_ouro_calibration.py": "e9099d1cddec60482c811df656e7aa8731c2a1cd98ae274efde249636d0cec50",
+      "loopQ/tests/test_packed_bundle.py": "cfffba2fd0d640e2e5def1d6375b24a87b4d33bb57a7c45fb5d08145ae46bd4d",
+      "loopQ/tests/test_packed_weight.py": "387fdb51e3bce2cbd7c0f94dcec5fb1ecd8eedd55c79b6811975503625be884f",
+      "loopQ/tests/test_packed_worker.py": "41206a263d72bdc2e7ba65c8ecf7e6982f3b31b62c9602e554798212f8373a37",
+      "loopQ/tests/test_paper_benchmark.py": "70eaa2326bf0b8da12736a1577421a42af446aacd48fc21be0e9ead2deab52ed",
+      "loopQ/tests/test_paper_math_reference.py": "a46176f310f90d75bef72674620a2425de13a20bb65f84d33686dc82ff4043fb",
+      "loopQ/tests/test_quantization.py": "a0368909de90a7b5d64d0cb82296ab438306f4e337c50ed1a103616fdd3f9af4",
+      "loopQ/tests/test_saved_tensors.py": "522eae339ca2d2b01506f653b18d6417bc04b4315d0d1afa90b78820d55c7742",
+      "loopQ/tests/test_sharing_gap.py": "aacbacadaee3850d1d823cf8122cbc2ac80f2af3066254a7b5b60d46c0ded1ac",
+      "loopQ/tests/test_trajectory_diagnostics.py": "341d8650a0c3ae59c12584baaccc6fec454bb76f988e533385f5953d3f83bf3e",
+      "loopQ/tests/test_transforms.py": "5e9fc5b387fb8f40f3b090864a0c3751529a869bef176058bac149dd4401bfca",
+      "loopQ/tests/test_vllm_bridge.py": "ccab13d1a9c3bd18279e46b4268cccba84e766153282e9d78cd58e093fe50647",
+      "src/loop_transformer_quant/__init__.py": "a9dceaa469212264a6ee7b1a615de2c5cd6e6195a007c852fc7e1e85857c172f",
+      "src/loop_transformer_quant/adapters/__init__.py": "6d112e61f7a6dc6093f11278d03a5351e8dd7adcb55d34dbb9e6810f86dbee74",
+      "src/loop_transformer_quant/adapters/vllm.py": "9f8df73df24a2b2297def5f0b994b8508fd06dc1a16184af64220e9016b813fd",
+      "src/loop_transformer_quant/cache_codec.py": "cf22a2cda1546e2bacb1e7e6ce9d8bdf8db952d758dd5cf81fdc1e784154185e",
+      "src/loop_transformer_quant/dynamic_lse.py": "ca0ce130dd1f838ab34c96ba8fe23f308ba09975801f587e4d5fa5723ddd93f5",
+      "src/loop_transformer_quant/gsm8k_calibration.py": "f90b5b89e595ef86ca6d03ae4f77bfa62b123aee3b9efcacfa468a87c72d64fc",
+      "src/loop_transformer_quant/huginn_anchor_mechanisms.py": "bb59f9f3d90ff0bf3dada6cd7d48ddf26169796a0553a4c2a39fd520c2222b8a",
+      "src/loop_transformer_quant/int2_experiments.py": "7be8a00a688a57ee7aa0e8fbfe6e8adf15277dcacabd95e608b5707d9010abba",
+      "src/loop_transformer_quant/int2_runtime.py": "2eb7f992fe80c63874efb440f8ed3b1c84374d16118872532d9ada3cde3a0b1e",
+      "src/loop_transformer_quant/int2_streaming.py": "e16f5ab4abd71af8e3ea5ec3cc3164a5956bacbe87cd81f1b7c2af551bb2bd75",
+      "src/loop_transformer_quant/kv_codecs.py": "118c98e0e32d899d7e6458d8c626738912af9ce87d761789a0d7d2a51e8d24c6",
+      "src/loop_transformer_quant/layer_channel_heatmaps.py": "04f09e794e6f87d75410a2ff55bbd60f047bdaf5355dcb4abca085b0a0fb318f",
+      "src/loop_transformer_quant/logical_storage.py": "01858edbf3c5406af60c747375c0fbee6b406fc867a0ea1741525caeb90245f3",
+      "src/loop_transformer_quant/native_activation.py": "4f82f90e80574e92e2dd97128a206983f32e0bb424286fad33ceee91c19dd366",
+      "src/loop_transformer_quant/native_kv/__init__.py": "acfa90ad9c90018df7f68d2efecfd8f9dbb4ca90344279ba3aa307e6829a5b39",
+      "src/loop_transformer_quant/native_kv/cache.py": "573b46adf72ad17415ad2e1a8af977cefa63f3958698fee7ecdc50b2a8a8ff2a",
+      "src/loop_transformer_quant/native_kv/calibration.py": "6b5123c4933e147f44c8a47c339311a034f842957155894f81eeea0f4be0e674",
+      "src/loop_transformer_quant/native_kv/channel_lse_certified.py": "43d0d6dcce1de9fff4c99979336a0861c67f7bf6d7ba786509201ca03e8cd0cf",
+      "src/loop_transformer_quant/native_kv/channel_lse_cost_probe.py": "8a816598f39847d32318383e456dab2e575e04047ed89434fcc830428ece23f8",
+      "src/loop_transformer_quant/native_kv/channel_lse_fastproducts.py": "c7b4b4927725c7388342aee92abf014304e81714c27050bce4709725a0a3eb19",
+      "src/loop_transformer_quant/native_kv/channel_lse_guarded.py": "04f2893e42049ed01b7c0c4a4076f9ae5ad786562fc81311e2ab6ffbdca0fefc",
+      "src/loop_transformer_quant/native_kv/channel_lse_kahan.py": "036991681f5e6667f6684c0b794eefdfdcbdffa2b096095413544a848c96ba00",
+      "src/loop_transformer_quant/native_kv/channel_lse_mixedstats.py": "3acb22432e21870c99d83f12121cec2882bff1ad9b14ebc27600699b68ea9cc5",
+      "src/loop_transformer_quant/native_kv/channel_lse_native.py": "fe7d3082aba93d483983e091053d722860b51b478e138f92a27114c5c6ee7a11",
+      "src/loop_transformer_quant/native_kv/channel_lse_pagemixed.py": "2b83812cf5133e4a88780d4a89fa7179ec1ea581f68b4f356551d896e162a981",
+      "src/loop_transformer_quant/native_kv/channel_lse_pagetiled.py": "5f38be7492b93176cbe998d3b33345056a6db90c40275261b5c65ce3631cf104",
+      "src/loop_transformer_quant/native_kv/channel_lse_reference.py": "b09e33d551bbe31a6fd56469a207f8a6c9173eb915761c3c263c2574ee31567e",
+      "src/loop_transformer_quant/native_kv/channel_lse_solve32.py": "eeafd3685da864208cffefef9d5bd3575a8a2b873baf7f38e79dcbcf3fbd64f9",
+      "src/loop_transformer_quant/native_kv/channel_lse_split_solve.py": "d847cccf05382f38c337c094654e892c1544e37dff1fbb89967e349220756ccf",
+      "src/loop_transformer_quant/native_kv/channel_packed.py": "083f30a6446e125079df2e17811d06d004ee289953b999304f95e989acd86513",
+      "src/loop_transformer_quant/native_kv/channel_paged_fused.py": "a7fc2ab879859804913838b38c7888db589420f2fcdf5303165ff6a72c4d6f57",
+      "src/loop_transformer_quant/native_kv/channel_paged_grouped.py": "e52cf976ee34bd680af8107b18e97e50b7272deac2b3925e5c12f1d248171ca7",
+      "src/loop_transformer_quant/native_kv/channel_paged_meta.py": "179c3df4189e034f5c5b233f23e76bc0a49d12bab990480525d5f033ecc2ea6e",
+      "src/loop_transformer_quant/native_kv/channel_paged_recompute.py": "0ca703afd01d66337c949af2b3d73513bf801cfecb41f1c7851120e7e6641f63",
+      "src/loop_transformer_quant/native_kv/channel_prefix_scan.py": "3ef7845af24a20f515b2f04461c97fb2eed3d9d9200cc60d77bca4b7b34b0c1b",
+      "src/loop_transformer_quant/native_kv/channel_split_write.py": "8b197f54626758564c220aba8908de298ae2f4929a7131abbdf2d9fdea192421",
+      "src/loop_transformer_quant/native_kv/channel_token_fused.py": "eb455108f066f1ad90821a9484efc2ac4967f1b72d5ac83773054ff36ab16f36",
+      "src/loop_transformer_quant/native_kv/channel_token_meta.py": "443f3332b6e31cba118e880f71fed429c007f6d53543625d273b66eef4fe90a8",
+      "src/loop_transformer_quant/native_kv/comparison.py": "d318411d27db8f903ca1f250db8196d3b490e7892f9ee49d647dc3f304f88b00",
+      "src/loop_transformer_quant/native_kv/dynamic_paged.py": "c34a21557bd218278b6a08a8cdaa0b0a30a2b29a1ff02db02579f9cb7c73556b",
+      "src/loop_transformer_quant/native_kv/dynamic_paged_fused.py": "3c0f5c620c6bcb938bea32a94aa832c70fb356716088dfda207f47004727e393",
+      "src/loop_transformer_quant/native_kv/dynamic_scalar.py": "69e954ac3100cddb7464ec3e6caf9c3d3927fc96b99c472868057aa821a69867",
+      "src/loop_transformer_quant/native_kv/fa_tile.py": "5836fc96d54b5892fa911f0d0d48f81c9a91d71690c0455be639f31958bd77d0",
+      "src/loop_transformer_quant/native_kv/kernels.py": "d5537c45945724076984af5ff68c9e7b859528bf5c6ccc437c3885fc96f002f0",
+      "src/loop_transformer_quant/native_kv/ouro_comparison.py": "bf232a254102672863fe233e627a4e0059286a7f55e6362e397b22ce83178e77",
+      "src/loop_transformer_quant/native_kv/request_slots.py": "662be79c9ca67d8dace1c843a4ccb1b9325cb76b0b97c9c1fcdc677e44e86e25",
+      "src/loop_transformer_quant/native_kv/static_paged_fused.py": "19dfc2d58beac8b5318a97ca8fa2b5ac3badc4f33c0ea013d70e9db4d33d8466",
+      "src/loop_transformer_quant/native_kv/xqa_loader.py": "4de9c6c9f3a68d4fc422d68d3e55868dc756686299f901127acf284148472b0e",
+      "src/loop_transformer_quant/nvfp4.py": "68f0623c31f9667b295d09b2ecf4832aee84dbf93ab3de7348579189ca44a7d1",
+      "src/loop_transformer_quant/packed_affine_kv.py": "3c249789ca964aa1947262e9177fc3f66401d67889d1396a511cc489ea89e682",
+      "src/loop_transformer_quant/packed_rotated_hnd.py": "844421a98211c9a4bc1867c79f19c7d5583e82abe606b941276260c9a9e9c662",
+      "src/loop_transformer_quant/packed_rotated_hnd_safe.py": "9409fd0cca643ed8b3c87de99440e66eb7d3e9e8d6fdce8c76553806c3c29323",
+      "src/loop_transformer_quant/packed_rotated_kv.py": "195427fea9b110b693fbb54a1cd809482bdec7c75417055ec9ad14c8ed182ecb",
+      "src/loop_transformer_quant/packed_rotated_table1.py": "9b246daf90bf8d005482d413fd4f72422945d8ac001e98c61075365a6e1e8885",
+      "src/loop_transformer_quant/paths.py": "1f87fce7344da5916fd4a8b88dec2f5d7b45bf4b4cab129b72e9bf2280595d9c",
+      "src/loop_transformer_quant/pre_sinq.py": "e7926059546171473b26d1fdcfb2378884fd95aefe573d93119acf981e13240e",
+      "src/loop_transformer_quant/quant_policy.py": "bc55de3e7d27d7825cb2dbf443cd90ab3fe0452856a38864ecdb93c2e0fe3e38",
+      "src/loop_transformer_quant/rotated_attention_adaptive.py": "6ca61bf1207abe12e40de81900348df49d1832a69d1d2b9b92fee81daceedc34",
+      "src/loop_transformer_quant/rotated_attention_adaptive_safe.py": "5c93b4f091f723a1631b210bfb0589501c191d87adc57eacc8e90c98ae3b3f2c",
+      "src/loop_transformer_quant/rotated_attention_combined.py": "24d0fea4c099321594fe0d80c82093e934fd231137559d1a9d90535470db0458",
+      "src/loop_transformer_quant/rotated_attention_combined2.py": "207860a760e4a2bb2491d233247fabe965c84796ad7eb1bdab83df9a64d3d77c",
+      "src/loop_transformer_quant/rotated_attention_combined4.py": "df49a846ac8059bbdd9e5441fa91d366005063d0946c1c1fafff929998b5ca06",
+      "src/loop_transformer_quant/rotated_attention_conservative.py": "6318dbf853294db80b397985b0f3b92fce7d97791c752a0fb9df949095b4a294",
+      "src/loop_transformer_quant/rotated_attention_fast.py": "2e884106998c22e14fe822d3267478da54521ebfb93ab76e174daec23913ba9f",
+      "src/loop_transformer_quant/rotated_attention_group_all.py": "d6973ecff1f66eed99f9955b102f6ba5d51d240e249f724d6667028998da0347",
+      "src/loop_transformer_quant/rotated_attention_group_codes.py": "be0723b591b7c95436057c3ea7616c4178e516c8dc80495b698375196239e921",
+      "src/loop_transformer_quant/rotated_attention_group_codes_v2.py": "327119e221287a8221d178290aba72c4d428ff1d6664347374f04d287146da83",
+      "src/loop_transformer_quant/rotated_attention_group_load.py": "7cdda56e000afe900d7b6d3a942f112c4ab986a1a70476275531dc3238696063",
+      "src/loop_transformer_quant/rotated_attention_group_pipeline.py": "b2139be31958bb37fdf0afbfdfdbcde8ab91245208782ba16631ea47b369f3f3",
+      "src/loop_transformer_quant/rotated_attention_group_rotunroll.py": "f4bdc70394069dc078649b6ad5310ccbce2e5d9dff048f403fcebba687ce6fb5",
+      "src/loop_transformer_quant/rotated_attention_hadamard.py": "2dc3ea5984f251fb17d93c4868cdc6297801723bc6799ac1a9d4d96bc18e46e2",
+      "src/loop_transformer_quant/rotated_attention_pipeline2.py": "81efddc54396b60dd7c2c412e092e43eddacb0d9641f2d8a4eec9878d72eb4ee",
+      "src/loop_transformer_quant/rotated_attention_query16.py": "161faf38eba65c083fac344853dfade1ec326da7d6257469d10626c2abce5df1",
+      "src/loop_transformer_quant/rotated_attention_query8.py": "143350254d30170998b5300874e1c04a44adc5e3cf23563b33fe61dd1e8c77c7",
+      "src/loop_transformer_quant/rotated_attention_rot_keep.py": "2f55a8ed07a478067ae68b911ee457b4f0628beeae078b121eaae0c8cf5d14e8",
+      "src/loop_transformer_quant/rotated_attention_rot_transpose.py": "dd7823da1e66c88a4f7b74874d25d94261cbc17e85fb02a73938fcfce6fb7765",
+      "src/loop_transformer_quant/rotated_attention_unroll2.py": "90bb8c89664acff921d0369083aa7f53ed628b0168ff5d2e8779654fe727ea97",
+      "src/loop_transformer_quant/rotated_attention_unroll4.py": "4cd2474e5a63dd893d3fc545d12fdf0ea2d2be0e4d232bed1130b5aa7ce95539",
+      "src/loop_transformer_quant/rotated_attention_warps16.py": "c984459e4fc0d9d1d3d0bfe8ec41f1abb7e9ad723da30506ba44a68ca25085f7",
+      "src/loop_transformer_quant/rotated_attention_warps4.py": "44401f179631c6e6120b2e111326e77c14a2edd002ec28d3f8168d102d22695e",
+      "src/loop_transformer_quant/rotated_attention_xqa.py": "e7a8a71abac60a1865a092d25777a499b4fbb6ba8722430c0d4e705f28f4f786",
+      "src/loop_transformer_quant/rotation_policy.py": "7fb06183b276662564c2f64f14bd7ea6a5438b46b3d8a05efc3678f46c4b54cb",
+      "src/loop_transformer_quant/rotation_study.py": "c05e433e8182e7af43ae499d09fc5eb5f8a53512cef4d01be9dd2d59ea26e14e",
+      "src/loop_transformer_quant/rotation_table1.py": "c7bbd247eab272600bcba2219c83160bb54abc4bcff3d1de8ca496fe2faff985",
+      "src/loop_transformer_quant/runtime/__init__.py": "c412e782654be12e77250c6b617b5034c9a2e33a1084b7880f8c0eacb46e05f8",
+      "src/loop_transformer_quant/runtime/cache.py": "2e1f0026738ab93ee3767ca29a53ff38a67d707f3e64f466fc402a80eebc82fa",
+      "src/loop_transformer_quant/runtime/context.py": "e8f723bc270d1293b086d621c66f69a964e570175df733e23c9755e74933ca10",
+      "src/loop_transformer_quant/runtime/module_cache.py": "f189187bbbc1436049ba863550ac0b4134466e09bc1ee3471c3b030b33d01e17",
+      "src/loop_transformer_quant/runtime/specs.py": "b9a3deee69e44c8ccc009e92bdf34c83662b7d553ffebe25eabbead0b897bdfc",
+      "src/loop_transformer_quant/sensitivity.py": "00d0ab71278e19113dd6b5db65d091271fe1be65054562677edf7c082d8f3022",
+      "src/loop_transformer_quant/sm120_fp4.py": "301b7a8141337b68fda40dd9fdeba2d4eac46fbb17c0d61292481910ece83ca4",
+      "scripts/account_recurrent_kv_storage.py": "00aca9766fa7960c010b5170f36781224e45a4a288518bf37c6864b14422fbde",
+      "scripts/analyze_accuracy_grid.py": "0d33123eae6c5beaf625854f6d0b92e4169a835e8fb067139c0b037dcfcf2e46",
+      "scripts/analyze_huginn_anchor_ablation.py": "3d825b07b42a08a8f9dbb8fe7b39bb50c8b51c5c7f02da8183ad11a90d43906a",
+      "scripts/analyze_huginn_full_validation.py": "6b8a5972ff7f878d2368273a41898ccbd5321e61056a59c8ac845fd2f01a7741",
+      "scripts/analyze_huginn_mixed_precision.py": "d8cd6ee6c8f39ec28fffb6ebbdc618443bc88918dd1c5ebe06a8ebac30289519",
+      "scripts/analyze_huginn_residual_kv.py": "015e3bd1abbdc14c98b077d645609ef4fc312654a8c854f421a62e0516d467df",
+      "scripts/analyze_int8_scale_underflow.py": "181ea6a32f88a9edaab237edaa5fcb02d5baa85d518ce996a53ffba214601388",
+      "scripts/analyze_int_scale_matrix.py": "b0d17fc01e36e0cae0bb2fcd12b67e578b0b67983159ee5dbf582cd01a5c7b12",
+      "scripts/analyze_kv_codec_screening.py": "c712a196068fa991a2272435d5c302edd570fd4100209dcd5bf316103557ba35",
+      "scripts/analyze_loop_delta_quantization.py": "dd1201e93c2183a428b1bd0b514a782169d81d03cbec2fed82db67e80f3932e9",
+      "scripts/analyze_loop_kv_redundancy.py": "872e86904e189d18e8440b7f79e80bcbc0061d1b9780aee26c29459710c9d7cd",
+      "scripts/analyze_ouro_selected_wa_kv.py": "f2da633a5e23fa92d9c444b80553149b8644257f479641ef2eec8a9dffd88c3e",
+      "scripts/analyze_paired_screening.py": "777b63be6a0c0b27a07115f94f2da7f50ae8da36818413e6a7b00044895fd79d",
+      "scripts/analyze_predictors_vs_accuracy.py": "0d0c94a94b5c81ab2b11700bbe6a9ceb4d2802ca240d7e1d274954e4ccbee0a0",
+      "scripts/analyze_recurrence_context.py": "2765504af81b1d9025374c1af39d11914977ae1067090292e5a9ecee76a98dc7",
+      "scripts/analyze_run_repeatability.py": "1d529562e36480fa64d32902303c74080f23b5f3358cf04c4e611742df359b46",
+      "scripts/analyze_selective_w4a.py": "ff84fce91a09e11f1d7351a19268086ad9d493e8bdd0657d200fa0128683dff4",
+      "scripts/analyze_sensitivity.py": "5a1d93a6fe3e261f1fe9d54f5535904c378a23b13ef39f482dcea90056c77f0d",
+      "scripts/audit_activation_table10.py": "df38f4deccb5cc437ffe8537a5364277920743c10cdcdede17b1b26709e4e796",
+      "scripts/audit_channel_capacity_reproduction.py": "14558f10129915128726e7535a6f67929e1e71dfc9132865fd94c595ade0b2ad",
+      "scripts/audit_fused_write_model.py": "4bd0788465be0a16fa54709760a5a6ede71d46b1c8128a0f0601bc4bcbb2932d",
+      "scripts/audit_huginn_table5.py": "93913720f049190fa1fcebf93cec2f26d2c802775bb71a5424bd35d227902a6a",
+      "scripts/audit_int2_accuracy_smokes.py": "c0a463d3cd3fea5889c1d758999a3cc059b43c89a57c229513397de6c4521434",
+      "scripts/audit_kv_cost_final.py": "c193155a1d34f2638c05fe65d23b8d68349ceea7586a2acb0f88a8f7cf547abd",
+      "scripts/audit_native_wa_method_screen.py": "05cdf99ca7abbb7333ddce355ca93db4d642b3187fd332007cc5d39bc816d2c0",
+      "scripts/audit_ouro_table5.py": "2966ba1479831d7c941093b5709c00f4b29a7f39a1d8d095025244470cbb35d6",
+      "scripts/audit_portability.py": "465c903e2a1581cc1f84ed7883cecc524295924c31b72dfc685d973460539909",
+      "scripts/audit_quantization_goal.py": "3362e9fc93d2ad3d2aaa89421b8e187c74dd284c6c4a60ad71fe8101505d2086",
+      "scripts/audit_recurrent_speed_goal.py": "7410c511e4ca51c15afedd744b913a2e68fee2b84abab4968e130261ff13bc21",
+      "scripts/audit_scan_g4_model.py": "83acb9495b81dc3ec6d7160c6aceda110826f32ba8f1a50ae1cf5b24d84180cd",
+      "scripts/audit_scan_model.py": "bdd572caf68eef4fd634ce14841512aeb5e2bc64f907cdecbb682d08f0c3a3f3",
+      "scripts/audit_split_solve_full.py": "333caecc946276e9986c9d0a1bb68ab388b4b2b8bbe32b7c1090378cbdd48e33",
+      "scripts/audit_stream_solve_model.py": "11e43ee2ac2f2116684f2818735d980e00613052a06ab53b78993ea92c1cb442",
+      "scripts/benchmark_activation_workspace.py": "1f3c65b83e6b09e211335f0d5710ed9d00533500706a96f36b4bf803cc42c157",
+      "scripts/benchmark_channel_lse_costs.py": "ed77ceb5d1aafc2b1102b874266b6c5bc9e07be8e7ce3e1af14b3a95f0a3fe1f",
+      "scripts/benchmark_channel_lse_policy.py": "362517508242ed94d83e3b11024115be2cd86b71efc7dd4112cbcf0b053a921d",
+      "scripts/benchmark_channel_packed.py": "64be08868ef380c3b905044026db94c289edb1dba849d7aaf74f15d4da6b2d78",
+      "scripts/benchmark_dynamic_lse.py": "0642a3cbb797c802d45bceb60e4b270ea0e793041020320b3489eeff7e14eb71",
+      "scripts/benchmark_dynamic_paged_write.py": "b607ff8de2d0ff083a963cc71d26c7a4d515999e50ebd576353a657d4efbde75",
+      "scripts/benchmark_huginn_native_components.py": "f39093b1d1d339e7fc571db16b2c8c4662b616fd2691a42f35518f16595b4fa8",
+      "scripts/benchmark_kv_cost_probe.py": "6da5ecf0c9767ce784934ee2bee7a0852dd8905a50a921e212520a4d6cfb74c1",
+      "scripts/benchmark_native_activation.py": "ec82e0f1c126b525d49e785f3e085a2ebe769056e77a1a2d13c0aaadc1c755e5",
+      "scripts/benchmark_ouro_native_attention.py": "c5a04ba7798b3631a189187e30bcbfaa08423508f671f08979fa071c445b7364",
+      "scripts/benchmark_ouro_vllm_fixed_generation.py": "92a3d379a73d12350d2c1df1afd98d65146205c412b902bb0051b9a976c4ef9b",
+      "scripts/benchmark_ouro_xqa_seq_trace.py": "d7fa0a4be2e6940e2b0e86071eadbd95004d89b69fe854b05add2fc65cef92b0",
+      "scripts/benchmark_recurrent_kv_chain.py": "085af93b3250eb492823166d7d304ded2ed050dfb42352df80c562c9757f122c",
+      "scripts/benchmark_xqa_candidates.py": "7b130c23368b584b06d3d64ee6dcebd102eff7064ce64edd109bd08563daf4ae",
+      "scripts/benchmark_xqa_costs.py": "ae0b11a31900d21b1e143f5d88fd36887eedcf977154f9a0890870d5fa8fdb6d",
+      "scripts/benchmark_xqa_native_attention.py": "7198b554ed0c10d7d2681c2c297bce03fd085651b552404e4790448d6e8f3cf9",
+      "scripts/build_bf16_anchor_reader.py": "9c072bb017932ae62c2885fcc942c98f8dd7f9e29299151d35ce495e70fc17b1",
+      "scripts/build_completion_manifest.py": "84b8ccc2eafb27025997b981bc9237404cc85ab36f01765101be788eda274369",
+      "scripts/build_fa_tile_attention.py": "5ff13df779225048a1b8f42de46e6a99993569edd51725ab2e3067f0f1bd6ccd",
+      "scripts/build_m1_storage_matrix.py": "a5eeaa9eaeca126e03be047ee4537478f068f5a730616bee8a36558a9f41be86",
+      "scripts/build_provenance_manifest.py": "96eeb4ec1c23af52a5e4ffb415aac532356eb37370c862d5176a4053c100646a",
+      "scripts/build_recurrent_attention.py": "0896c6280513fe2b624c16451e8773007e2acf685adbafb467f89a9127c7f950",
+      "scripts/build_recurrent_read_view.py": "def6ccd89fafa3db068972e426c348e9d29c67d2a4f3506529e3b44c0509d981",
+      "scripts/build_recurrent_xqa.py": "a87ce431cd27ac95476ca6382677fb28093f5942a4a87ab9c7f2d057adbb0a9f",
+      "scripts/build_vector_decode.py": "2eba31bc992c7895b1c17b8588590001617e9707ae2ffc8621f82e572b56369c",
+      "scripts/build_xqa_candidates.py": "4e0e4cab2c5927025f5946a5a79f73f43c3386012ed5b4b95741cd63eb0c7ffd",
+      "scripts/build_xqa_comparison.py": "8fbdad07ddd5ac35df2acf14a5a01a8522eea83877b7cb35ec6a553cd72bccb4",
+      "scripts/build_xqa_cost_variants.py": "88d90a75f931a205dcbb13079846aba305a542e58b0447242194ee39c8b8b5a3",
+      "scripts/build_xqa_hnd_native.py": "a3e0ed25569036b194b0fcfb6f59b8478145396e351cbb0db05dc83ffa2b48c2",
+      "scripts/calibrate_huginn_residual_activation.py": "7189b032c99b5f5c25d2349481cff3f97a4a919ad9510c8e925f97315c7c25b2",
+      "scripts/calibrate_huginn_residual_kv.py": "a5809dae0f75d5c7e45b2453ff1576e3bbf05c1a86f6fcb17fcca03bb69da0c0",
+      "scripts/calibrate_native_recurrent_kv.py": "de867b19016e5cf3d90fac9b1684f13874bebdcb7658e34aff1a059403a0cca3",
+      "scripts/calibrate_ouro_rotatekv.py": "f995a50d5b35adda2560c1f26ea44482cd78145968a07336fbd06868ce50e553",
+      "scripts/calibrate_residual_activation.py": "0c5348f5a0989bf4f178c877b637dae3d9a3a1225e28986760aca39002c5a600",
+      "scripts/calibrate_residual_kv.py": "1b7d6c1c23bbd251dcb7ce3f9f1dd994b581e6c8d8237a0f7c9fa79c950ff9d2",
+      "scripts/capture_int_scale_matrix.py": "f507a18d3394d0b39e94bcbbb45e27cfedb68a9435a3b9fe27079cc365065bc1",
+      "scripts/capture_ouro_kv_reuse.py": "72f411ddcc7e546874aea70e54cee946ddce7e6f85c558087dfe1e2fe7a940da",
+      "scripts/check_activation_workspace_cuda_graph.py": "e0d2c9b871949286a6b0c45b8e5a8b713c6f8f0ad2dbfbfac033084308be2310",
+      "scripts/check_bf16_anchor_reader.py": "107caee9b7a3a58d1554016feca042c545240ce5c4f2685f5edce0d1d783c6d0",
+      "scripts/check_channel_lse_certified.py": "141e493ef9ac943fd65e4b0352c48844ab1e250ee577064d98b4790364ca93c5",
+      "scripts/check_channel_lse_native.py": "7937d8c0d3b94830984ee8fa19c9af7829689ce4ff1ca20f448276d587fe993a",
+      "scripts/check_channel_lse_split_solve.py": "ed222515a14b35c617f80259d91d53b98563a2928009f6126d7a7b93844b1734",
+      "scripts/check_channel_lse_split_state.py": "8634d6c172652760f57070b2ae9a782b3c1fb865832744591f3b89ee2272790b",
+      "scripts/check_channel_mixedstats.py": "eb80353d62f2569b012589290510a165d4a9d04190f2e15c319534c366401b32",
+      "scripts/check_channel_packed.py": "7d5983fd3850838f6da437196f29459f91248cf223900c063f5032ed37195094",
+      "scripts/check_channel_packed_heads.py": "4cbd129906f2ce7a5f4e91eafb50724bf49eda9df4eac126531fbdb80b342c1a",
+      "scripts/check_channel_packed_split_solve.py": "75900d0a014901182629864c5575ee0d01bfe4e1b92b3dc387535b26560973fa",
+      "scripts/check_channel_paged_fused.py": "72bbadb9214c5815050f1c3436659a6619632c42d6bcf10d4cadca8665a7719d",
+      "scripts/check_channel_request_reset.py": "eb785b1301ce8380eb7052b4433c63891f7d4bd14a369125af1930ce2a5ae4eb",
+      "scripts/check_channel_scan_numeric.py": "11c090307331ed7078c0986e3c5436bf1b5387f078afe78320bb9e645d5d429c",
+      "scripts/check_comparison_attention.py": "fb8b20d36f57efd3eae00db870471e285eafcc68bfaf91ab9a7a54d119aa0571",
+      "scripts/check_dynamic_lse_recurrence.py": "7d6ed664c3004ff6445ccf9f6d3d7c4501fe346a6e5b13a745e159c0855e669f",
+      "scripts/check_dynamic_native_scalar.py": "fcefa1178f06cd7281c9fb67c858bf70b6857afce5699b3617f11b89a97bdda7",
+      "scripts/check_dynamic_paged.py": "b42960a5f7795f13646e98acc937b810b9f1e12cc1b9ba0d8a0d00a3feae12bc",
+      "scripts/check_dynamic_paged_fused.py": "cea6ff572d72bbfb4f03a5b0325eeb6bf17758c58f052487d4b40c3eec1ba456",
+      "scripts/check_dynamic_paged_heads.py": "c17a61e56ba883cb92b9aa54a5d8b557abd337af32cf9711b90979f7dc0b042a",
+      "scripts/check_environment.py": "00343fbd026557b2c317c26883c37d0218495512cb38bb99506ecb1c744d5490",
+      "scripts/check_fa_tile_attention.py": "1b23aa2a0d25c3176c175c8e841e616e112ff40e70a4c1e4178ca881e629d9e2",
+      "scripts/check_fp8_read_view.py": "5d2cf0e70bde90177c5a6be7234c6bb94fe303957e595ff2f7ea1858cab49708",
+      "scripts/check_huginn_graph_parity.py": "e1b5002ed78f7ba49a23e62db86f29841365265330c5d5cf3bcfcc87391d8695",
+      "scripts/check_huginn_sm120_fp4_parity.py": "6916213489c904c13994ae663b7f219c3890d45c9247353600e66929d3c3462e",
+      "scripts/check_int2_experiments_gpu.py": "400ceb0731884a208d3b5225fb4994e682a92a0163ec854a308c55c9fd62b091",
+      "scripts/check_int2_streaming_gpu.py": "dbaeb741a1215f06de00b78bc0f71c6e6c01b8a100ad9c180b4ca0f6848c2b5c",
+      "scripts/check_math_modelopt.py": "7924917242ed769f80ce00d18043ef7b719706c480259182c076f98ba270b62b",
+      "scripts/check_packed_affine_kv.py": "acc10f3ddaa6b38957c42f0741301668700ad2882711117fa990d574feb53b20",
+      "scripts/check_packed_fp8_conversion.py": "b05e9ef6f321715f93341d079c4a0c13cd1c516763d91be4b55957ca02016314",
+      "scripts/check_paged_comparison_attention.py": "c35a2f8983149403f6f13986ad38e4e3df331908cbee5ffafb70eeca75268359",
+      "scripts/check_recurrent_attention.py": "18d513c8b2545f7c0f393b5edbdff0451fef22bebb6f2a3b6ca7c90083090179",
+      "scripts/check_recurrent_dequant.py": "3717d5bb5d012a73e3c51416330d82310e6baa81c7777a32277920a78b890e2b",
+      "scripts/check_static_paged_fused.py": "69d122d8d6bd49728990ea71a5437a1ddca2ce8a2d6e01e3e15cf067f1372e55",
+      "scripts/check_xqa_bf16_matched.py": "34a5a6469befd2f372caad2b278d1ff1d9b0b06b842e63003c36c81927acbe40",
+      "scripts/check_xqa_fp4_lookup.py": "5aab5b1a773ef3495e65e69cbeb36657d884b1cd0567c831285c91f747edffe4",
+      "scripts/check_xqa_fused_prepare.py": "2fb54c3928dc77e369887517396bad325ea44ec019c2ec29bca9c0f85925831b",
+      "scripts/check_xqa_half_product.py": "b55b1632d22a562271815e55250f8f6ca1493ae8251f1fcfbe56e14d1eb8e61b",
+      "scripts/check_xqa_mixed_attention.py": "b0708b1cf369377d5e65efd637b2126bef1e33d5be7940e0b2d7fd6f990a1de4",
+      "scripts/check_xqa_paged_attention.py": "9d0fad47bb905da01d8c0eb2d76084dedac321bf9771fc349de6ec42b346d870",
+      "scripts/check_xqa_paged_generic.py": "f3efb933095fd47da14c64e86d463dbfeb4e7731c5e0dd50cec2a6bc8f2ae15c",
+      "scripts/check_xqa_prefill_decode.py": "bcb51bd5947ccc1ca2c490ca8a17dcfdbf1c68312a6a441b724ddfd1998184db",
+      "scripts/collect_activation_momentum_payload_heatmaps.py": "e773e670b76f5b5a0af600edc340535f028121d122daf2bb3d5c50c13aaca2d0",
+      "scripts/collect_gradient_tensor_norms.py": "a1d6d20ae81e700badf807e445a770a24a2f2dd2cb50c7659dc07efe910fa031",
+      "scripts/collect_huginn_anchor_mechanism_cell.py": "b9c58fd7fc3fd6caa8ee5875f4a8619ec736646afbafcc4156285e15d0caa055",
+      "scripts/collect_huginn_layer_channel_heatmaps.py": "87977e4d6a34e3422fa42af1ac943cbd17814eb6911f8f19297318b72ddeeb6a",
+      "scripts/collect_layer_channel_heatmaps.py": "41a293cede5636d07ab52baa61d7c495b63cfde5915ad9daa8a15db62aea99d7",
+      "scripts/collect_momentum_payload_heatmap.py": "0b94a57d71fb7b8318c21f9f2e6fc2e604317debd441a1a73fc5d0dd76c28c9e",
+      "scripts/collect_ouro_gradients.py": "d7b7c0671906c75c353ff34b257b60ccd5405479647f85bdf957204834fbf493",
+      "scripts/continue_after_q08_uniform.py": "10c0b22748aa4066bd69c49f9dc4e9c4ae50068fef2e1bd7b953bce4568412eb",
+      "scripts/diagnose_channel_prefix_cost.py": "f29704a8cc5d3fd3ce81892fcb06395349d15ed64a32d41f2936be8a18121e6c",
+      "scripts/diagnose_int8_scale_underflow.py": "a5c134d657da16dde0cab5eeb09fcbd3e3c2b837e8ea8447defe0354f6546b40",
+      "scripts/evaluate_gsm8k_cache_codecs.py": "2bf64cde6944cc6cf3ac3e88ef8d22336a27fc4da6c2d7b4505c0fd930a275cf",
+      "scripts/evaluate_gsm8k_fake_quant.py": "22d0c237f74310ed2a8de0973b8bbbe59bee5c57ade8e5fa35f51b7e941d0b8b",
+      "scripts/evaluate_ouro_attention_precision.py": "8fc0f383cae171c45fb3029b83ff42bdf0e0e37b761c3e6e58abb4dd30ff9562",
+      "scripts/evaluate_ouro_code.py": "649415991ebaa7894c40e0900c76408ee2fbe27a6a483f5438cbe2d1d99712ff",
+      "scripts/evaluate_ouro_recurrent_vllm.py": "d0f14e137f425dc1e0f6d2341dbefab3f4df2d04a313760ae99e2b2ff80bee2b",
+      "scripts/evaluate_ouro_vllm_attention_precision.py": "b945f987d6dcc98a25f8fd2f1eb307774e3d4b6cb1170c5e2b2d981fa54b9ba9",
+      "scripts/experiment_manifest_common.py": "49b5ca26f2087394144ddaa6d0a36b07f16e503b7a838a382fd7b8f2257b74e7",
+      "scripts/experiment_queue.py": "21df177893db181b87b13dfc04d9e5538e81cb4770cec8225df7485d91c0badf",
+      "scripts/export_codex_sessions.py": "212d5d0428d31aca43345cf6017ebce23dbfa98134ac85ab0b1a8ed16f2fb0b0",
+      "scripts/export_native_nvfp4_checkpoint.py": "b7373493e7a13586827c370f8521e544dcf238a16792c89ac60174c8a7eb77e9",
+      "scripts/finalize_activation_table10_report.py": "4f9bb32ce670eb61f270e9dfcfffb7786625c5e73c42b111e0c66d792a1eff41",
+      "scripts/finalize_huginn_table5_report.py": "dc1994ae56f9100a92c4bc5417c10bab9b7d148d81200732740acb2522ad17aa",
+      "scripts/finalize_loop1_dynamic_table15.py": "753e956cb404d2401cfca8928eaf5d8cd568a9d17c1b93971a10a5ba2e83123c",
+      "scripts/finalize_loop1_remaining_resumed.py": "94cdb86481d84ddcda7abcc7831f9b3a93da857cc41cf3e592453c8369fa224c",
+      "scripts/finalize_q08_uniform_bits.py": "3c653a6ac63e6495e1c293fdfaee85c302cd657d770457e0c63735cc5910c752",
+      "scripts/fit_int2_accuracy_candidates.py": "bb9d87115ac4213cff11d12e96d83e1e5c4eade0ac67821fb5b3df326c9361d7",
+      "scripts/generate_activation_table10_tex.py": "f8f855ba9870dafab119fe9ee3da4d085814f36270c1286296a2ea4d339a00ca",
+      "scripts/generate_huginn_table5_tex.py": "7253a086a2cc10630211307fec2142c7e12ec06e010e0c64662acdc0a845f11e",
+      "scripts/generate_kv_codec_report_assets.py": "9f0fcfd69147f1bc7feb3a54b9cbaa0c4465bd73eeb92b4d8f8fc99008dbb975",
+      "scripts/generate_notion_goal_results.py": "c24ce41bf220cef7f3bbcfe3b0b656a9fce58c28a444f71506f741e683ecbb3e",
+      "scripts/generate_nvfp4_residual_report_tables.py": "830a8a6c038f739f3dc44c056601db515fe82f3df94c118dbf625dfc190d8dc0",
+      "scripts/generate_precision_flow.py": "24b713b71784ec3c84f52940b535d1027bd5ca40fb73f81baaedbffdcf7a1bda",
+      "scripts/gpu_availability.py": "09c716b2fad2e969b3eda9acedf91817d4dfbf0d67280563bfd2588386a486d3",
+      "scripts/huginn_table5_matrix.py": "22eb8f679a832ecbbe6f2d43ddb9653c2103ec3417cbd246b56e33abd0e59325",
+      "scripts/measure_sinq_nvfp4_weights.py": "cc2b80b416b5c47b415bb4c60ece87d394ab61b751fb726c206abee9c7c11d58",
+      "scripts/merge_accuracy_results.py": "a8d0d79613893024382ed874b497fa08154bd1ffca88e3bf8ff669216ef378a0",
+      "scripts/merge_full_validation.py": "ea7182148e8b98defc87b664176570a75ab08f9bec5c05799e70c25b635cbf53",
+      "scripts/merge_gradient_tensor_norms.py": "850ecd33e3fe3539fc83b6adb8232d58a8e3f6a540ec1c07c4cc17ab6b38205d",
+      "scripts/merge_huginn_anchor_mechanisms.py": "8d02dce1f8914cfcfe18118378dd10c370d52b619e598bc181fb65d6e7b431fe",
+      "scripts/merge_kv_codec_full_validation.py": "579713ce35513f10537b2c3abba3f7b33aedc05d73aa76c01d103c85adae48cf",
+      "scripts/merge_layer_channel_heatmaps.py": "5a56b2407d5ada4ab61415e00f25c4780926d1ce2bffd03707576eda52d6cfd9",
+      "scripts/merge_sensitivity_shards.py": "1b1afa38007afe6bf4d4e54e211cbb956b3b6af79b1fe0a24300a33c99851e06",
+      "scripts/plot_delta_ce_summary_grids.py": "7029fe408fc3e2e8dd29407d22c2c6b8fa574b8a374827791704c71271282672",
+      "scripts/plot_gradient_tensor_norm_heatmap.py": "f888343b48a10cf142532c391768526cae058bce7a70197453dc1387715b0fc3",
+      "scripts/plot_gradient_tensor_norms_v2.py": "7a057b1c2b3a7251eb8ea565a64a32a82cad4bde00abd2b4c6ac53bb0d1ae00c",
+      "scripts/plot_layer_channel_delta_ce.py": "c7167a3a26b92e7804c4594494e0e102df57fb3293efd15a703d97e37e9c3c9b",
+      "scripts/plot_layer_channel_heatmap_grids.py": "f0ccfeb3f0dbaa6e3e7710b2e8e36a5dbded48565aeae1dd78565c1fafdb4d0c",
+      "scripts/plot_layer_channel_heatmaps.py": "e436e4e3b24945bac35785d0c2882ae4ef253d9ccb88281333230ba0275ef601",
+      "scripts/plot_quantized_payload_heatmaps.py": "8b652720d49a64b32336f09b3b7d72d21b754d6c84d238a821c612ea9328d37b",
+      "scripts/plot_quantized_site_gradient_heatmaps.py": "89859986a1625c6143a562eb6441c99760c7a310f287f95cf82aed0551a72ab3",
+      "scripts/prepare_huggingface_release.py": "a0ad51ea23de34f4c5023d61eb6afe23c9ac65a4c1d151c7ad574ff8c3614326",
+      "scripts/prepare_huginn_pre_sinq_checkpoint.py": "2ea485abd7d3f9458e243a768d8e4fd54ccd0bf23b2b98489608703533a1bd2d",
+      "scripts/prepare_ouro_pre_sinq_checkpoint.py": "923838b1d95298e9ccaf81d11e4dd39c6f4a7834596e354075f7d5ee60253be4",
+      "scripts/probe_split_solve_vllm.py": "e8d8b3ac144dd8c9b8574f3c0db35f706a1f97a1153bc3b7bacda4a14fe04f8b",
+      "scripts/reproduce.py": "7479a7e865d62d16d245745efc964e532eca419ded3aafda7d2a4951cec88697",
+      "scripts/rotation_study/after_hnd_memcheck.py": "bbc9751a541c33a3f96bc2d3b5d14657e924b3207ba96a21613ffa69b9797c11",
+      "scripts/rotation_study/after_microbenchmark.py": "e0ca715d17ffb68bb61671a19250d546b2165be15ca903db70a2448b8b1685b4",
+      "scripts/rotation_study/after_resource_audit.py": "06f2c32b4f0cdc6ac7dc91e1f7b529c731cb3695af97459b9171326898b5d8a9",
+      "scripts/rotation_study/assemble.py": "59f0f0fe62e17f54b63deef52bc8c6db3173d05d781632a055f58f5f2271e642",
+      "scripts/rotation_study/audit_adaptive_safe_resources.py": "e9c9a60429e83b33391aa9451ca452ed2a0b0f10b998c498ea580903c2f47256",
+      "scripts/rotation_study/audit_combined2_resources.py": "949ea01ea47dada23e561352ad10fdbca1bcad40febe89043b39a73e6d9f9e10",
+      "scripts/rotation_study/audit_combined4_resources.py": "5974bf8fe837f653f01e72aaeb19e5a5b3d5da82e3e131db136e5e1fc699b0bd",
+      "scripts/rotation_study/audit_extended_resources.py": "23522c6deab12c048be5398e26d13e0286dd58d337165e4e82bf129d5883e1c5",
+      "scripts/rotation_study/audit_extended_resources_v2.py": "c1d7d7d2dc185726c60a5fddfa54e98da9ac59fbd146a7956e03a323c79a14c4",
+      "scripts/rotation_study/audit_kernel_resources.py": "0e75d8bc8c926167f2bbb80b929aa2a0ed1987652757ea5735062411b11f10e5",
+      "scripts/rotation_study/benchmark_attention_conservative.py": "0738cf41aad38302ccef140f84a5ff4c7369811f77ce10f8a65dbf538d7f6a54",
+      "scripts/rotation_study/benchmark_attention_fast.py": "7ab5657d2d65600f4123764fc2fa68710a5bcee7ea9d48315a159f582166181a",
+      "scripts/rotation_study/benchmark_attention_group_load.py": "7731b1c4ba41a4c1233d9affde6ff390adc52e8694cba34d58191e6d90c2ed67",
+      "scripts/rotation_study/benchmark_attention_hadamard.py": "c9dfcf998b9966f24be6bb28beca5860b1c5f3e3eef2d39c13c7aaa17e08b76a",
+      "scripts/rotation_study/benchmark_attention_pipeline2.py": "72ade1025ea6b7a7cd75c16b9dc511288c0457af1f17650ed9a28483bd7c568f",
+      "scripts/rotation_study/benchmark_attention_query16.py": "50ec0df96debf7efcb79be22794d9224a41ef8a6661ed1934fcf700dbcc9e4dd",
+      "scripts/rotation_study/benchmark_attention_query8.py": "b694221a3443dc5678eef6fee96001451d1b0b6b8a8dec4c13d6353c23459e9a",
+      "scripts/rotation_study/benchmark_attention_unroll2.py": "a33454010fd4476c29668f9b8244303dffc6e1f809c4fd1d84b8648d0b944058",
+      "scripts/rotation_study/benchmark_attention_unroll4.py": "75aeba671bd83fbb51b9ccb53afffe3cd5a33522653ddc488974de44ebc1d5b4",
+      "scripts/rotation_study/benchmark_attention_warps4.py": "b9b5ae3e8e6e99c6749624da82a13cec9242289438f6fd9cb2f60f71ba050f95",
+      "scripts/rotation_study/calibrate.py": "229bd2299d8d2fc581c9070788e63b75b991e15fb1d713952fecde6a3f888dad",
+      "scripts/rotation_study/check_attention_combined.py": "bfdaa9f9f05064053372ae3cecd3b111817676ddea88ae4d7a50a46adb74b9fc",
+      "scripts/rotation_study/check_attention_combined2.py": "102fc9d461c92391debc06443f045c863f935ef3526b0edb8d99c413c25ae17a",
+      "scripts/rotation_study/check_attention_conservative.py": "3c618c8acf3761c91fc03162f76209f92875b909bad12f9d240892150eecef33",
+      "scripts/rotation_study/check_attention_fast.py": "960716ba5ccc9fea0daf7c2b5b7807862a30a1ca6190096969b4ec98cba88a88",
+      "scripts/rotation_study/check_attention_graph_growth.py": "e13e721cf4c77e319832f079d37d84c0499cb8096f1a8885a2dd4224b2b5cf3b",
+      "scripts/rotation_study/check_attention_group_all.py": "903fb7ce4ef652b9a83b3f7dc670b7eccd81a4c4c27db4c036e7f35f7c561df8",
+      "scripts/rotation_study/check_attention_group_codes_v2.py": "1c7adde1dfa15bda45403f9e3280c97bc41c2917c41cb3a4b09e1e8b57bab1f3",
+      "scripts/rotation_study/check_attention_group_load.py": "e0df7bb939e31299536af8711816defb19e4d6c0ed9922d9ad3d5338c0cbb6dd",
+      "scripts/rotation_study/check_attention_query16.py": "7234778ff9fcfa4682cb8c473f2aeda2596cf05564b5c5108d9a0f8028a5a058",
+      "scripts/rotation_study/check_attention_query8.py": "67cc348b035731dce563728a7c502d980ebff9c218ec87b06c8a082921f1415a",
+      "scripts/rotation_study/check_attention_xqa.py": "f1ea76d15c447ae926daef4532c86de9f74d14c9e2c7ec6ab731439869ef20e1",
+      "scripts/rotation_study/check_clipping.py": "456503eb2ceb13231f896e6bca8bae707d0fb8c0c34f3bfbaa88813089044766",
+      "scripts/rotation_study/check_packed.py": "75ce77cc5689f0d580418631b1b7d651db88d67cc2214af80b56f1d6cf839bd4",
+      "scripts/rotation_study/check_packed_extended.py": "a61f4df471eb030553b5c3845773743ca4ca90e956550d4267bac289aa8f70a9",
+      "scripts/rotation_study/check_reference.py": "96a1fa527711697f34ad9f9b9721da53e842cde85f9799c64203a022721fd328",
+      "scripts/rotation_study/collect.py": "175fb6af33c49bd09ef5df3a6329aebc97a2666c5b605442125cc930f1eed766",
+      "scripts/rotation_study/diagnose_hnd_alignment.py": "f49d2ff32017adeb6ed60bf718f7397e74e83bea0d15b5f064fb8d58112f3339",
+      "scripts/rotation_study/diagnose_hnd_alignment_safe.py": "f7c3e64765b0cc6b86b4e374e46485c92d20fef33bd6a28984bb1a7e1172d7db",
+      "scripts/rotation_study/evaluate.py": "32621676094335cf3d96eb81801d72288d9d6390a7329f9acf9c06a3aabf3a01",
+      "scripts/rotation_study/evaluate_combined.py": "1d755d82819038b94c3376ae6f9c06b6cc8e33119f8aff3722ea4de9bc8fd973",
+      "scripts/rotation_study/evaluate_combined2.py": "683300fe00fd9a2bef03c7bce257ef7a290e032e901b5bc90fcc5fc92f1f45ab",
+      "scripts/rotation_study/evaluate_conservative.py": "fbcbb6cc82cbfe1e16515b3f5f068ab356ad214ee37bfbab7adca15da39eccf7",
+      "scripts/rotation_study/evaluate_fast.py": "4a6ccfac980e651f6813deffcd8ea14d435186cf29c4386e3ba4041b169f434d",
+      "scripts/rotation_study/evaluate_group_all.py": "ed9a980dfd7b2140107ba2d0748940df49cf636af0d85abb804b82838c946216",
+      "scripts/rotation_study/evaluate_group_codes_v2.py": "e5e24323924dd11083681d5a9f74ceec40677463ba9704ab8d479554dbf530fb",
+      "scripts/rotation_study/evaluate_group_load.py": "6b5f72d640017a8787d4f4c00d4a155158b662e073cc68f3585de7a8a6d35d25",
+      "scripts/rotation_study/evaluate_query16.py": "cb3d9282444f2c2a388ef16a7875a88c69c88f8a4bd4314bd91212e2124970ec",
+      "scripts/rotation_study/evaluate_query8.py": "b14553e09b497cbc9b64b7f12ee1d8770ae1db01d1efadc1f1afd0ba39810e30",
+      "scripts/rotation_study/evaluate_xqa.py": "447268526c0b2a146a59a0ebfb81fefa5e7120050322f95e2ca212ac1bca3074",
+      "scripts/rotation_study/finalize_recovered_study.py": "0659a2c8cdf10e7f297356a6f3cae744f63c6f6494fe9167314946c2c77470e5",
+      "scripts/rotation_study/finalize_when_done.py": "0f97ba9270afb4782d699815c0afca65e462a363fd0f626f0733bc8a81287e90",
+      "scripts/rotation_study/pipeline.py": "3e79d4d7fbea491ea361def4ce86c2077871151bc8ac8b4d2c44664ac3eaad14",
+      "scripts/rotation_study/profile_attention_costs.py": "91ec37ff0c5e316c68aa9f449f696da41e9e1facebad4a6e03f61954b7e4d6fa",
+      "scripts/rotation_study/resume_full_cells.py": "fe0a80d87a046343119fe3fc0771ba8597235d1879ce8ff0f60102fd1f87bd42",
+      "scripts/rotation_study/run_acceleration_comparison.py": "2edb6c1f4fb5d6d7900b0a26ab3e4a1890f8fb030527648952299b8cc62d5771",
+      "scripts/rotation_study/run_combined2_comparison.py": "f344b81b7e96f7db40a49a2f8f52da685f699e9005f43df78da7a5a2a1d1be94",
+      "scripts/rotation_study/run_combined2_followup.py": "d403d837d6d62d4d6c50fc0c3e06ecc28c273bc0842966c3a312dbdf38915cbb",
+      "scripts/rotation_study/run_combined_comparison.py": "a4a276513fa311a44488d0fdade035916d5a8a238ab41a619d7152b8a6585c42",
+      "scripts/rotation_study/run_combined_validation.py": "c8d22d1f36189ba7d8b77b2284b5bc1d1b71815d4496ac0e4b92eb1681db5bd1",
+      "scripts/rotation_study/run_conservative_comparison.py": "61405a1b15adc4d3c0e0c7db08897b98300d7cacd7183fbbb34a8da63db25412",
+      "scripts/rotation_study/run_expanded_acceleration.py": "e89890efc49cbc49f544c80e6e82fcc1acca7032bdad038a5e466e9ce8fa0f1e",
+      "scripts/rotation_study/run_group_all_comparison.py": "6dd53bc538e3b377535ef83bbc344e64db37a3d7ebcb4cb6e276f9f1a6251186",
+      "scripts/rotation_study/run_group_all_followup.py": "e42d4e2230db6a02f589d0d8a5d9dfdb09a24a2c5d49cc0a063f88407730a3bb",
+      "scripts/rotation_study/run_group_codes_v2_comparison.py": "129de34774e3191999e2c75fea8990531b4a8ae9ed8392d935710ac18214e177",
+      "scripts/rotation_study/run_group_codes_v2_followup.py": "457030747991052eb77fb3152cbdca78584abdcbbd1e5cfff98d22450c0090ff",
+      "scripts/rotation_study/run_group_load_comparison.py": "61e770792b1eb50f5bc76cc2f8368611839fe51353e84b89366b07935fa215d0",
+      "scripts/rotation_study/run_group_load_followup.py": "c23dfc84017bbe3d7f84e418bb51e505335d732907327a1ccf2a44f9e1387af7",
+      "scripts/rotation_study/run_loader_screen_queue.py": "825bbda7fbc6549ccf470ca958349110c0487d43a752871cd70a4a563dab710e",
+      "scripts/rotation_study/run_query16_comparison.py": "ffcaac8aad3c9db92567e4bef208a484dfb4063781f8732d3ae16f7c581ee9f9",
+      "scripts/rotation_study/run_query16_followup.py": "e8a7efda0d5a50e7e87d939e90cb755485ef4975a2295a4710157ace4284ffab",
+      "scripts/rotation_study/run_query8_comparison.py": "594aecbcf665813567d0f5c61c675504da6cd77f241f597f856f5f60e500516d",
+      "scripts/rotation_study/run_xqa_comparison.py": "ecdb552092650d2cba17a36e8fbe736d67c5a08737f15183ee3faf5f55ec942f",
+      "scripts/rotation_study/screen_adaptive.py": "c47cb8699940ab6ed5c822ec203d73545dfd290d1bbf01d9657ff8cf9ecd0a79",
+      "scripts/rotation_study/screen_adaptive_safe.py": "d40993ff4e40a9a617f54c35165ef52c6d3c58146e8117f112992b3433aa469a",
+      "scripts/rotation_study/screen_combined.py": "bbca232a238a01f9bf54252723980dbbe0c132d684e551334262be4f956f9fae",
+      "scripts/rotation_study/screen_combined2.py": "b35609da38f021ce247f7cab965e8dfa0ea7a2be9abaaa1c8175cd9e1f5bb840",
+      "scripts/rotation_study/screen_combined4.py": "7912e85e5dd2cd10ba2b87e4099ec5d50e1b7229f0ad024f0fe3f5afe6005b59",
+      "scripts/rotation_study/screen_group_prefill.py": "8b50b8a773174e37a1467db8cc7cb9d25e9263f8fc0239ccdc4097c8f8cb360e",
+      "scripts/rotation_study/screen_hnd_cache.py": "d208cfcfd036325d7d1a8b406480fc1f432f3833172b26426bbf124d0d34b6f6",
+      "scripts/rotation_study/screen_hnd_cache_safe.py": "f7251a4dbffeda1f465bccbd48f9ee201e10d461edb0998c748abdc596752eb3",
+      "scripts/rotation_study/screen_loader_candidates.py": "d201f41d4bc5f5c5f33f2fc62c601087f2eef629f1e1e3e4a87a4732891ed3b8",
+      "scripts/rotation_study/screen_loader_candidates_v2.py": "c2aee9b56fc05510f38880f2072a12b65efc938d67d5fe391cdd6a9a8277edf6",
+      "scripts/rotation_study/screen_rotation_access.py": "5cd4cbd0df0aaf9d7c8e154bd4a206ed5f1f9b47155659dafaa87641d4d71a54",
+      "scripts/rotation_study/summarize.py": "54be9e99168c4cf430cb27ba10de54fb6d6e35268b4466c9b45b0ee63fd246b3",
+      "scripts/rotation_study/summarize_expanded_acceleration.py": "1d15778f3b1e093ebfd8b4deb8fccbbe99035752ca691f2b0d0be445a28ac933",
+      "scripts/rotation_study/validate_acceleration_finalist.py": "1432e674519f118fb26ead5ba5d1396b80b3cb42e4f55b95be6d90a4c6a389a6",
+      "scripts/rotation_study/wait_validation.py": "26265d2d904c2734be7fdc2a518ede8b51493d22d7785b332d32c0d77f383aa5",
+      "scripts/rotation_table1/assemble.py": "85022a1d7ebd6758206162ef7eb0420b580e2c33c826db908892e8198231b680",
+      "scripts/rotation_table1/assemble_identity.py": "07c311f0589b47010f958add0defbc9b789af481d92e99098259cedd744bf528",
+      "scripts/rotation_table1/audit_smokes.py": "537a099a365dc36f616059887e280d369c363d4088aaec88780e77755e0fa412",
+      "scripts/rotation_table1/calibrate.py": "bbf40471f159f7af193362271aa4d2a7b6ce34e8a32959abc8fc7983edfb0b1d",
+      "scripts/rotation_table1/check_math.py": "ff0b97813f57eaec01035859d178c5b8aeb9828ae34008acd32731fc363e7fec",
+      "scripts/rotation_table1/check_packed.py": "57aa4980c083806bed498a2025f3e4f868cb03a6e35867f4cc722d4dd7bde961",
+      "scripts/rotation_table1/evaluate.py": "37b81a409d5bf8b9f5449839022d98059fb90c851c19c7a19c32f4599a7ae902",
+      "scripts/rotation_table1/plan_table3.py": "d6f579cab190f6442f5d6fd1b74f61d5244a1a78698eeaa76e120e01b5267212",
+      "scripts/rotation_table1/render_table1.py": "eb9d6e548fea8f753f11475e80146b4cc7b535a76a24c2d67bca9fdca42660b3",
+      "scripts/rotation_table1/render_table3.py": "5bc33b107c92760cfd6c3dae3b0dd7e8e72cb1e68533b17a44208c82081b40ab",
+      "scripts/rotation_table1/run_full.py": "b4bdcf7fb41d4e79525bd146a0fc589792f721769cb6441e7a17a799d4c475e2",
+      "scripts/rotation_table1/run_smokes.py": "d216bf8354b3c9d18bc90ba103a9095a39b80c8c20acab27656ce1d7828fbf5b",
+      "scripts/rotation_table1/run_table3.py": "1ee705fa75413c69fd0ee11a44029cf2c1bb511c36cc9858f2cd65bb35d1f258",
+      "scripts/rotation_table1/select_method.py": "c1d2f45d1f7bff4639e9c12211889076289483176a84435e3a9dadc7a46f572b",
+      "scripts/rotation_table1/summarize_calibration.py": "4b5bb9bd7d07cba440da7ce89bf365de32ee8b6d2843aa23dc4969bbb85d479c",
+      "scripts/rotation_table1/summarize_full.py": "bf94aac5c5ccc07cbfd82e38f01834e6989f386301f45fc8f378e177f688f81a",
+      "scripts/rotation_table1/summarize_table3.py": "1323e4a7cb40fc13862d1b455b241f241608cb151b72dfa629363f540cd843f3",
+      "scripts/rotation_table1/table3_jobs.py": "c368ef161c9bef456a39d215673f7b0f98ae75cddc8363cbf1b3ce1dcf533faa",
+      "scripts/run_activation_predictor_matrix.py": "5e9756e04612221cea145990d118443c3e747d516bd1ee3ce21d4e0403ca763b",
+      "scripts/run_activation_table10.py": "be359602b5e147ddd0d28a8f28726dbd9b857622bcc037bc7843e91caee992bb",
+      "scripts/run_coarse_sensitivity.py": "f327fbb2e2b80cb98d6b00b49e96484dbce1f479bc67d4e33b1b6cc9ff2d16f8",
+      "scripts/run_dynamic_group_scale_multi.py": "69d91faf446b9103d4c00f5175ec1a0f786b5e719bcaf3c9f94872dc87bf01ed",
+      "scripts/run_dynamic_group_scale_sweep.py": "4e8588a8f1ca7f02b566d8ede5373a18cf56a014d84487d458bcf4b7c2d64b68",
+      "scripts/run_fixed4_suite.py": "46eacb9ddfa89466cff33a1610d2051e53dd22a2489c933008757c1ba3eb8605",
+      "scripts/run_gsm8k.py": "594e570e58ea79aaaf27e6fa20088e4a63856044cf292b3d8aec57e658ea31ae",
+      "scripts/run_huginn_anchor_ablation.py": "d249e21f49f7df8421006fb847b51ad4848725ec0b7c16f4e2601bfb3599793e",
+      "scripts/run_huginn_anchor_ablation_queue.py": "544e241d06cb2aa34ab7a5946c453ef184f64fda5567689989b2a2537dfb3b65",
+      "scripts/run_huginn_anchor_calibrations.py": "fd4f8909b3fdc08745a5f26021096f52999fbed7cf510f06a6c4135acb0907ce",
+      "scripts/run_huginn_anchor_mechanism_queue.py": "b201ca02b2f18849bba4c05103fe5323d428cae4bbd919c8db268c1bcee4847e",
+      "scripts/run_huginn_full_finalists.py": "c6d79cc94e4d2314c7499f785cc196ed2f31cbfdf12d19ed18fcad8704bdb941",
+      "scripts/run_huginn_heatmap_queue.py": "8958642e837aecd51f0eb1f5d05b873974983ca24fd02779e7a03cf94a914a34",
+      "scripts/run_huginn_mixed_precision_matrix.py": "f64249ff45a0fc0a1813bbc0960bd02b1f98c87e7571dfd697d79822f1b91286",
+      "scripts/run_huginn_presinq_native_screen.py": "592d3a35f85274ff3ef15386dc348f300185409c27c5673c81d54b525a9998d2",
+      "scripts/run_huginn_prespecified_full_queue.py": "266936101d144add1e56290205e01c4cd44e82441e303334402a388b1ef11a07",
+      "scripts/run_huginn_quant_suite.py": "77b3d88fa71b5ccb3710f844da84b2c0e1a87f572dff09ca90c2e311cfeff0a5",
+      "scripts/run_huginn_residual_kv_matrix.py": "3fc23607e30ca800eec5b4040d715a2746f3f3011e627d5cab9329315185e387",
+      "scripts/run_huginn_selective_w4_screen.py": "74f7c15da787e1b7a4c001c5ecac9ffdb6f5f1c9fb62da6d4c1cef1cf5197c7f",
+      "scripts/run_huginn_selective_w4a_matrix.py": "f1ee23f016b8704caaba0d09502e1a6cad237f5b99c912ca2c44242d87d7fd86",
+      "scripts/run_huginn_table5_queue.py": "cbd005233a00c9dc56b8b866845236a47867cc505e2009f47d8d89e914996a7a",
+      "scripts/run_huginn_wa_matrix.py": "1fcc70489a2cdc0a6db05c515b47699e0d4fe49939ab5f0fd8999df41f264502",
+      "scripts/run_int2_accuracy_full.py": "ace6af465d793dd1f33a922fc0de8ba8ac832c519f3f61dc91992b079671be93",
+      "scripts/run_int2_accuracy_smokes.py": "47c987f932ead87cd8715fed7df8c09f6bc18c9385c5d31d5f654c1023e5f87e",
+      "scripts/run_int_kv_table14_15_matrix.py": "b57b985ada939389206cb0d8137f1701216d609a3d231bfb1892c2b95b7434e7",
+      "scripts/run_kv_calibration_robustness.py": "4af1ca5de75b41ac1dc9b9c8967d13993434de4048080cff192859edbba49471",
+      "scripts/run_kv_granularity_matrix.py": "39f394a99e14d43277ba060daf70357038ce42d7075304b1f09149b68158c3f2",
+      "scripts/run_kv_layer_sensitivity.py": "619a4a422338dd2bf7bf892f1dbab6550e79ac9abb16754c9c27fc19ba6e75ea",
+      "scripts/run_layer_channel_delta_ce.py": "cd28a506b86db4d5c717f934e16328d37932c7a0b785b4c6e8613a6453ee73ef",
+      "scripts/run_lm_eval_timed.py": "acdcfdf1b9ffd040e0fadaa45b1942e3d1b548433f135fbe1c84e2fcad0cb6fa",
+      "scripts/run_loop1_remaining_resumed.py": "56b1f88d04e774e4a383097b11640ab3f91bf214fc7b7daaed0286e789ddcc95",
+      "scripts/run_mixed_precision_finalists.py": "0da37cf7c46b974cc1e0835c6ddf8b7dac31b0e50a00b3afec11501e6379e190",
+      "scripts/run_mixed_precision_kv_matrix.py": "4c3dba3d879add89b8b726dfc667d51e2432f78f8f3f283e5b467aa930ec1043",
+      "scripts/run_momentum_ablation_matrix.py": "8d6469daef62830027c2c937e62ac396cdb186af05f330eea4a5a907ae1c6fd8",
+      "scripts/run_native_nvfp4_method_screen.py": "3f8714578306432462d60e6b73b04905bb68433dd6fa7e4ed916dcb56103c28f",
+      "scripts/run_native_wa_bf16_screen.py": "fb102e51d601a442b61afd582e321e5949eac2cfaa0c07fb7ddda53c10e44cc3",
+      "scripts/run_native_wa_export_queue.py": "6abac94be9ddb192cdd306dbf13398502b6f88ee14047027cc71a8d0befa2de3",
+      "scripts/run_ouro_chrome_trace_queue.py": "549b3ddf9a776b2235d704bcf097e725e9c083860f5bd97748937222a8150fa8",
+      "scripts/run_ouro_remaining_queue.py": "d0e698329cc1e986941a062bab9c49b33d43e6fc069b94036c41918ad42d82fb",
+      "scripts/run_ouro_repeatability_queue.py": "3c44894a3b9ef6feb7d91c24d9360dc214ef1bcf320ccd93717a0b27247de05b",
+      "scripts/run_ouro_rotation_math500.py": "3e36126efec9527380e822aaa25cda4caaa1f98595c67abaa3c759892a34e7fb",
+      "scripts/run_ouro_selected_kv.py": "2186aa2dd115633af04fa6b32f996d0edbf4eca8dfe98dd9f60b0219e0b3f159",
+      "scripts/run_ouro_selective_w4_matrix.py": "59de40bab4307007ea068d1ac9a1390c8fb08156bdd6217f19a286cabd4d0b76",
+      "scripts/run_ouro_selective_w4a_matrix.py": "cd59845df8008f51137ee40093b152375afa138e949ff3ced49c752090b3dad4",
+      "scripts/run_ouro_table5_queue.py": "f102ac7637e4652e10da8e7ddc6c5e68228488fefa024841778bdee184fbfaf7",
+      "scripts/run_ouro_vllm_gsm8k.py": "8bdce3830b7e84376b343fda3bec6be9a1efa60a38527e9607d39c2acb30edb6",
+      "scripts/run_packed_affine_sweep.py": "eb8fd503a5aaf83ed3c236aa0ba580fd67e152ec2689e0d868e9f02a54db9b4a",
+      "scripts/run_presinq_native_screen.py": "77d9e4f4ab52e6c155b87afc20e7c674e145b0637ef7053fac0f42dc54123168",
+      "scripts/run_q08_offset_sweep.py": "595e1b0a8a1752a88f7ab3d5a1f9e2ede337768248ae93aa9ec9d6f6d0082d44",
+      "scripts/run_q08_uniform_bits.py": "990ee1bc1c1b5fc5f56d8b787ace89748eaf24645868d6c014a66e47d0b9b259",
+      "scripts/run_residual_activation_matrix.py": "02a21358423ea5ccb2ec83fc3b047073f641494bb61bb6b62837e648ee0693cf",
+      "scripts/run_residual_kv_matrix.py": "392e763fc90e4ea7e3b664e8f70d009a441c0301e87ac52ae5ec033d79013bfc",
+      "scripts/run_selected_wa_kv_validation.py": "3373b6d950c61a7ae8b5af144e01852cfc519e616e4a77416fa955d1f4e3c3f7",
+      "scripts/run_sinq_matrix.py": "3b87d45fb75b69c59a18a9bcb6efde414089671e1c4d579902719cac86a4a6c5",
+      "scripts/run_table10_calibration_queue.py": "e527825b7356cbf5a5d02e0581a8165b38fee0e93874937b48476bbe1cc8179f",
+      "scripts/run_top3_combined_kv.py": "bebfd392c92c744491502ff7fc59b74bbb51dd4be5cf74cc5146c1b4efb5de56",
+      "scripts/select_heatmap_delta_ce_cells.py": "a16a13f8c20b352759769389fb1690cc45b3243cc49824121ee35e313f4cf794",
+      "scripts/select_kv_codec_full_policies.py": "84c5655c547b3faf6052e998c6791ea41105ecfdf5d7137a53757259a75c1991",
+      "scripts/smoke_ouro_nvfp4_vllm.py": "4d6f608d2b9027a6b5f9071511fdef2aaca37e6cad917ec3862c5cf6cc1b3fa6",
+      "scripts/smoke_ouro_recurrent_vllm.py": "bb159fa91c75084a1ff0ae1fdaf6fd65f29a01e6b49d79a16e06c9aec98df749",
+      "scripts/summarize_channel_profile_costs.py": "1835ebab6097d373c9fcab3ccfc0208c2eb392a6c425d1093bb6f5b704b70998",
+      "scripts/summarize_dynamic_lse.py": "66f32fba2b6031ddc80224194a509ec9eadbaebc06ee7832e06c70acfbb1c6e8",
+      "scripts/summarize_graph_experiments.py": "0877b04fb282eb7afafdfcd7e6453b92e6ffbb3521544e7c0719bd87c748f275",
+      "scripts/summarize_int_scale_matrix.py": "4c937809ee46b2576e1aed96996a5525fd2798bcb9b8f83e65eef3dcf0bccb74",
+      "scripts/summarize_kv_cost_analysis.py": "cd409cfad8ecebf0d8d8a3aac1c829342da463f0a36a3ae7ccf63774fef0c563",
+      "scripts/summarize_ouro_attention_precision.py": "a0bf777fc366cc198629c2c467d0fd1f98335bbcca6ae876bb61fe9c68566319",
+      "scripts/summarize_ouro_chrome_traces.py": "47d7beb592fd8049fb61c3e82a0298b4f65f5c35890342f2cc97f73a56843632",
+      "scripts/summarize_ouro_fair_comparison.py": "d16891bdbbb74901cebda757b426e524e447c20bee7c87b36b8ab00bae5f791a",
+      "scripts/summarize_ouro_final_optimization.py": "5222fc60b20210f6786a66005dd32ff3f6a750e5bc92f67654ec325bf0d8f7c7",
+      "scripts/summarize_quantization_goal.py": "73dd960173630d82664d1cd240f71cb329d15fa9842674cd9ca1df115e1370c6",
+      "scripts/summarize_recurrent_full.py": "519bff759541a33cceb3c3513a9537d9272b9b0e8b9c58e3b9f6267bf8b0979a",
+      "scripts/summarize_recurrent_mixed.py": "406b963dbb92587f781402a6b386bdf55d98bf053aab4c0f25432c931510662c",
+      "scripts/summarize_recurrent_screen.py": "68fcc410a93a0d667b97e7ab2a334c55eaa1d7250a8d3ad8f245ce8988db7f9d",
+      "scripts/summarize_recurrent_xqa.py": "c4b3718a49fd016a7d1b7ef3d1772cd0f31ef038f5a596ea0eb63460e035bc62",
+      "scripts/summarize_runpod_benchmarks.py": "9d18c35c572aed7b4ca1ba8f08bd79db2094bcee61252733f7c3c1a19189cf5f",
+      "scripts/summarize_scan_full.py": "7c10989917660da80a64ae4414dd12bec35d9d8e3d24a0b0d6670081bc3fc234",
+      "scripts/summarize_xqa_exploration.py": "9abeef642993d443c99a5a98970c89688326c45a7245c0cad617442506664f4a",
+      "scripts/trace_channel_capacity_schedule.py": "6fdb5972552f69c96318764f1e50fda3d08487db3b471c7d56f8ebaafec905ba",
+      "scripts/validate_native_kv.py": "cb726f0a16755d968317cb133bf232de1b0135dfbc0fa2f73efab2557af5775f",
+      "scripts/validate_nvfp4_modelopt_hopper.py": "079cf32815542feb97b0784696bbac1488c57d562c70905cb67355e74160e8f8",
+      "scripts/validate_pre_sinq_checkpoint.py": "5f6a7d4daed934b88e30eb4dde79f62a272782b69a35a220f4899be966af8a2d",
+      "scripts/validate_reqat_e1m2_modelopt.py": "31a076de14ab0bd99facfa7a2ca19eac2aa947c2872fe64c70f08e87f40b0e0c",
+      "plugins/ouro_quant_vllm/pyproject.toml": "c16833fa241d6219d645f79979c5df2bc4a8c3494a74a35d4c8ac38d59e96fcc",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/__init__.py": "09961d7d247f2e2667eaa45747270f53172d32422e3a09391a3ad65d149822f8",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/channel_requests.py": "3389cf929bedb1dd3d4d38a7dec8ab598cba3fa05fef8746f3d32d9fa34c2474",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/fused_norm.py": "ea5b02eda1eed9280729ac2801ecefb8fc41fd82002bd76bf9ecce5c7c586e54",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/fused_rope_cache.py": "7765006ecd10e9b8562f5d479cff12f4471c807bab271954a699bb381dc97c83",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/kv_rotation.py": "b7d4ff4342d42a07d7791180c6ffad931a8d283ab0b46c7b1e4ae32865c0d436",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/loopq_bridge.py": "be35be25a58efdc2fb0ebc7f7101b01d9c8b0aebd66b0f0a60c21fe5557cf468",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/model.py": "806182c39a5693efd7895a607be17c44ccbc2b74d4e6624d90596c7a4cfb1f0e",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/native_attention.py": "f2ddc785386102a38a40cc8b108ab2c4bfb7fe99cd3dcbe2b400f747b59483aa",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/nvfp4_qdq.py": "b55c0270fdb6131848456141261d1c6b984f3f88ed31ac561ee7f981cb6a593c",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_affine_attention.py": "0a9a0a1018341df5261254fdc9fabf1e1534381d91ab7b91a8a7c8ef6e0bccc2",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_rotated_attention.py": "d8f7b9ec10f0c1c15e6043fc26444b4a6a376bfffdbcd8232e0ce577e270a01c",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_rotated_table1_attention.py": "7ae22f08c2efd240fb4c549943eaaae972a3161f482b91cbb5194bda3fd6b1a7",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/recurrent_attention.py": "2e6ff15630c0757a9a36ca14818535f7a605be2800de8083841493ba25524dc3",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/residual_activation.py": "8cca2e8ad00594220b2658c4be851efdc9627506c01900e8647c89463b8bc877",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/residual_kv.py": "4cad104ba295e294f87eb27eada1d968be46a7d5b4b0c08edb52e8bbebfc8e49",
+      "plugins/huginn_vllm_validated/pyproject.toml": "4538f596380c7d8487189accd14597cb2ed237dd76bb719a7057a95abefb4f14",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/__init__.py": "5e18f53c2fdb7911b6cc921decc552d1fa92369fc115e1a1199288e58bcb16cc",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/attention_bank.py": "8482b5175982303d7d999a5e0fdf71e2e6f09d5f34a88e57a6bdeb3df58a7dc8",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/grouped_attention.py": "1e7df3c0c3f7cebc40cb977c11926a0b2c032653c20454a160a7ecbc21fd7b63",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/loopq_bridge.py": "5cb3b60e24a2c683976d6229dd89135a04e3167ede696742a55a5dd327d49f7e",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/model.py": "252b5b87d06ee17896fc91945a645c1e528290c70be7e2129bf9f2271177e383",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/native_attention.py": "93843837ccd9b3393a080391d9c013a9586304af1acc1d850e52e8fbec7d2294",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/runtime.py": "c9b71f1f48d75475ea299eab01d5bcf6f38445a7c7eec62fe2d5205b47d6c218",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/wavefront.py": "a74f9b084a9e2708c1211dc9fb6cb20de23d61e17868e0967fcf0e6cd27eb125"
+    },
+    "scope": "repository source at launch; installed dependencies are identified separately by environment metadata",
+    "changed_during_run": [],
+    "source_tree_unchanged_at_exit": true
+  },
+  "result_path": "/workspace/loopq_data/curve_a4_s0/eval_final/hellaswag/harness/ByteDance__Ouro-1.4B/results_2026-09-17T00-06-20.554115.json",
+  "samples": {
+    "hellaswag": {
+      "original": 10042,
+      "effective": 10042
+    }
+  },
+  "metrics": {
+    "hellaswag": {
+      "name": "hellaswag",
+      "alias": "hellaswag",
+      "sample_len": 10042,
+      "acc,none": 0.5041824337781319,
+      "acc_stderr,none": 0.00498960683837106,
+      "acc_norm,none": 0.6809400517825135,
+      "acc_norm_stderr,none": 0.004651597209993087
+    }
+  },
+  "task_configs": {
+    "hellaswag": {
+      "task": "hellaswag",
+      "dataset_path": "Rowan/hellaswag",
+      "training_split": "train",
+      "validation_split": "validation",
+      "process_docs": "def process_docs(dataset: datasets.Dataset) -> datasets.Dataset:\n    def _process_doc(doc):\n        ctx = doc[\"ctx_a\"] + \" \" + doc[\"ctx_b\"].capitalize()\n        out_doc = {\n            \"query\": preprocess(doc[\"activity_label\"] + \": \" + ctx),\n            \"choices\": [preprocess(ending) for ending in doc[\"endings\"]],\n            \"gold\": int(doc[\"label\"]),\n        }\n        return out_doc\n\n    return dataset.map(_process_doc)\n",
+      "doc_to_text": "{{query}}",
+      "doc_to_target": "{{label}}",
+      "unsafe_code": false,
+      "doc_to_choice": "choices",
+      "description": "",
+      "target_delimiter": " ",
+      "fewshot_delimiter": "\n\n",
+      "fewshot_config": {
+        "sampler": "default",
+        "split": null,
+        "process_docs": "<function process_docs at 0x783c92691e40>",
+        "fewshot_indices": null,
+        "samples": null,
+        "doc_to_text": "{{query}}",
+        "doc_to_choice": "choices",
+        "doc_to_target": "{{label}}",
+        "gen_prefix": null,
+        "fewshot_delimiter": "\n\n",
+        "target_delimiter": " "
+      },
+      "num_fewshot": 0,
+      "metric_list": [
+        {
+          "metric": "acc",
+          "aggregation": "mean",
+          "higher_is_better": true
+        },
+        {
+          "metric": "acc_norm",
+          "aggregation": "mean",
+          "higher_is_better": true
+        }
+      ],
+      "output_type": "multiple_choice",
+      "repeats": 1,
+      "should_decontaminate": false,
+      "metadata": {
+        "version": 1.0,
+        "pretrained": "ByteDance/Ouro-1.4B",
+        "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+        "trust_remote_code": true,
+        "dtype": "bfloat16",
+        "tensor_parallel_size": 1,
+        "max_model_len": 4096,
+        "gpu_memory_utilization": 0.9,
+        "enforce_eager": true,
+        "max_num_seqs": 32,
+        "max_num_batched_tokens": 4096,
+        "enable_prefix_caching": false,
+        "seed": 0,
+        "config_source": "/root/venv-LT/lib/python3.11/site-packages/lm_eval/tasks/hellaswag/hellaswag.yaml"
+      }
+    }
+  }
+}

loopq_quantization/results/loopq_w4a4_completed/hellaswag/timing.json

@@ -0,0 +1,9 @@
+{
+  "model_init_seconds": 34.80067107779905,
+  "generate_until_seconds": 0.0,
+  "generated_samples": 0,
+  "generated_tokens": 0,
+  "generation_calls": 0,
+  "completed": true,
+  "process_seconds": 1380.8023844193667
+}

loopq_quantization/results/loopq_w4a4_completed/lambada_openai/evaluation.log

@@ -0,0 +1,235 @@
+INFO 09-16 23:35:38 [__init__.py:216] Automatically detected platform cuda.
+2026-09-16:23:35:46 INFO     [_cli.run:388] Selected Tasks: ['lambada_openai']
+2026-09-16:23:35:46 INFO     [evaluator:214] Setting random seed to 0 | Setting numpy seed to 0 | Setting torch manual seed to 0 | Setting fewshot manual seed to 0
+2026-09-16:23:35:46 INFO     [evaluator:239] Initializing vllm model, with arguments: {'pretrained': 'ByteDance/Ouro-1.4B', 'revision': 'e3b1e0993b1231a51d6069a870476dda4162c00a', 'trust_remote_code': True, 'dtype': 'bfloat16', 'tensor_parallel_size': 1, 'max_model_len': 4096, 'gpu_memory_utilization': 0.9, 'enforce_eager': True, 'max_num_seqs': 32, 'max_num_batched_tokens': 4096, 'enable_prefix_caching': False, 'seed': 0}
+2026-09-16:23:35:46 WARNING  [models.vllm_causallms:108] The `device` argument is ignored by the vLLM backend. To select which GPU(s) vLLM can use, set the `CUDA_VISIBLE_DEVICES` environment variable before running lm-eval.
+INFO 09-16 23:35:46 [utils.py:328] non-default args: {'trust_remote_code': True, 'dtype': 'bfloat16', 'seed': 0, 'max_model_len': 4096, 'enable_prefix_caching': False, 'max_num_batched_tokens': 4096, 'max_num_seqs': 32, 'disable_log_stats': True, 'revision': 'e3b1e0993b1231a51d6069a870476dda4162c00a', 'enforce_eager': True, 'model': 'ByteDance/Ouro-1.4B'}
+The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+INFO 09-16 23:35:55 [__init__.py:742] Resolved architecture: OuroForCausalLM
+`torch_dtype` is deprecated! Use `dtype` instead!
+INFO 09-16 23:35:55 [__init__.py:1815] Using max model len 4096
+INFO 09-16 23:35:55 [scheduler.py:222] Chunked prefill is enabled with max_num_batched_tokens=4096.
+INFO 09-16 23:35:55 [__init__.py:3400] Cudagraph is disabled under eager mode
+(EngineCore_DP0 pid=84955) INFO 09-16 23:35:56 [core.py:654] Waiting for init message from front-end.
+(EngineCore_DP0 pid=84955) INFO 09-16 23:35:56 [core.py:76] Initializing a V1 LLM engine (v0.10.2) with config: model='ByteDance/Ouro-1.4B', speculative_config=None, tokenizer='ByteDance/Ouro-1.4B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=e3b1e0993b1231a51d6069a870476dda4162c00a, tokenizer_revision=e3b1e0993b1231a51d6069a870476dda4162c00a, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=4096, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=True, kv_cache_dtype=auto, device_config=cuda, decoding_config=DecodingConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_backend=''), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None), seed=0, served_model_name=ByteDance/Ouro-1.4B, enable_prefix_caching=False, chunked_prefill_enabled=True, use_async_output_proc=True, pooler_config=None, compilation_config={"level":0,"debug_dump_path":"","cache_dir":"","backend":"","custom_ops":[],"splitting_ops":null,"use_inductor":true,"compile_sizes":[],"inductor_compile_config":{"enable_auto_functionalized_v2":false},"inductor_passes":{},"cudagraph_mode":0,"use_cudagraph":true,"cudagraph_num_of_warmups":0,"cudagraph_capture_sizes":[],"cudagraph_copy_inputs":false,"full_cuda_graph":false,"pass_config":{},"max_capture_size":0,"local_cache_dir":null}
+[W916 23:35:59.708204195 ProcessGroupNCCL.cpp:981] Warning: TORCH_NCCL_AVOID_RECORD_STREAMS is the default now, this environment variable is thus deprecated. (function operator())
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+(EngineCore_DP0 pid=84955) INFO 09-16 23:35:59 [parallel_state.py:1165] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=84955) WARNING 09-16 23:35:59 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer.
+(EngineCore_DP0 pid=84955) INFO 09-16 23:35:59 [gpu_model_runner.py:2338] Starting to load model ByteDance/Ouro-1.4B...
+(EngineCore_DP0 pid=84955) INFO 09-16 23:35:59 [gpu_model_runner.py:2370] Loading model from scratch...
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:01 [cuda.py:362] Using Flash Attention backend on V1 engine.
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:01 [weight_utils.py:348] Using model weights format ['*.safetensors']
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:01 [weight_utils.py:406] No model.safetensors.index.json found in remote.
+(EngineCore_DP0 pid=84955) 
+(EngineCore_DP0 pid=84955) 
+(EngineCore_DP0 pid=84955) 
+(EngineCore_DP0 pid=84955) 
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:12 [default_loader.py:268] Loading weights took 10.31 seconds
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:12 [gpu_model_runner.py:2392] Model loading took 3.0510 GiB and 12.135673 seconds
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:15 [gpu_worker.py:298] Available KV cache memory: 35.87 GiB
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:16 [kv_cache_utils.py:864] GPU KV cache size: 48,960 tokens
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:16 [kv_cache_utils.py:868] Maximum concurrency for 4,096 tokens per request: 11.95x
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:16 [gpu_worker.py:391] Free memory on device (44.12/44.42 GiB) on startup. Desired GPU memory utilization is (0.9, 39.98 GiB). Actual usage is 3.05 GiB for weight, 1.01 GiB for peak activation, 0.05 GiB for non-torch memory, and 0.0 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=38355684249` to fit into requested memory, or `--kv-cache-memory=42796003840` to fully utilize gpu memory. Current kv cache memory in use is 38512970649 bytes.
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:16 [core.py:218] init engine (profile, create kv cache, warmup model) took 3.83 seconds
+(EngineCore_DP0 pid=84955) INFO 09-16 23:36:16 [__init__.py:3400] Cudagraph is disabled under eager mode
+INFO 09-16 23:36:17 [llm.py:295] Supported_tasks: ['generate']
+INFO 09-16 23:36:17 [__init__.py:36] No IOProcessor plugins requested by the model
+2026-09-16:23:36:25 INFO     [evaluator_utils:446] Selected tasks:
+2026-09-16:23:36:25 INFO     [evaluator_utils:480] Task: lambada_openai (lambada/lambada_openai.yaml)
+2026-09-16:23:36:25 WARNING  [evaluator:335] Overwriting default num_fewshot of lambada_openai from None to 0
+2026-09-16:23:36:25 INFO     [api.task:312] Building contexts for lambada_openai on rank 0...
+
  0%|          | 0/5153 [00:00<?, ?it/s]
  1%|          | 50/5153 [00:00<00:10, 490.82it/s]
  2%|▏         | 101/5153 [00:00<00:10, 497.82it/s]
  3%|▎         | 152/5153 [00:00<00:09, 500.69it/s]
  4%|▍         | 203/5153 [00:00<00:09, 502.06it/s]
  5%|▍         | 254/5153 [00:00<00:09, 499.23it/s]
  6%|▌         | 305/5153 [00:00<00:09, 499.51it/s]
  7%|▋         | 356/5153 [00:00<00:09, 500.74it/s]
  8%|▊         | 407/5153 [00:00<00:09, 499.62it/s]
  9%|▉         | 457/5153 [00:00<00:09, 497.53it/s]
 10%|▉         | 507/5153 [00:01<00:09, 498.14it/s]
 11%|█         | 557/5153 [00:01<00:09, 492.71it/s]
 12%|█▏        | 607/5153 [00:01<00:09, 493.65it/s]
 13%|█▎        | 657/5153 [00:01<00:09, 492.21it/s]
 14%|█▎        | 708/5153 [00:01<00:08, 495.75it/s]
 15%|█▍        | 758/5153 [00:01<00:08, 493.75it/s]
 16%|█▌        | 809/5153 [00:01<00:08, 495.88it/s]
 17%|█▋        | 859/5153 [00:01<00:08, 494.76it/s]
 18%|█▊        | 909/5153 [00:01<00:08, 494.87it/s]
 19%|█▊        | 959/5153 [00:01<00:08, 494.79it/s]
 20%|█▉        | 1009/5153 [00:02<00:08, 494.73it/s]
 21%|██        | 1059/5153 [00:02<00:08, 496.18it/s]
 22%|██▏       | 1109/5153 [00:02<00:08, 496.28it/s]
 22%|██▏       | 1159/5153 [00:02<00:08, 495.78it/s]
 23%|██▎       | 1209/5153 [00:02<00:07, 495.97it/s]
 24%|██▍       | 1259/5153 [00:02<00:07, 495.68it/s]
 25%|██▌       | 1309/5153 [00:02<00:07, 495.36it/s]
 26%|██▋       | 1359/5153 [00:02<00:07, 495.89it/s]
 27%|██▋       | 1409/5153 [00:02<00:07, 494.94it/s]
 28%|██▊       | 1459/5153 [00:02<00:07, 495.31it/s]
 29%|██▉       | 1509/5153 [00:03<00:07, 495.78it/s]
 30%|███       | 1559/5153 [00:03<00:07, 496.21it/s]
 31%|███       | 1609/5153 [00:03<00:07, 497.17it/s]
 32%|███▏      | 1659/5153 [00:03<00:07, 497.79it/s]
 33%|███▎      | 1709/5153 [00:03<00:06, 498.41it/s]
 34%|███▍      | 1759/5153 [00:03<00:06, 498.75it/s]
 35%|███▌      | 1810/5153 [00:03<00:06, 499.47it/s]
 36%|███▌      | 1861/5153 [00:03<00:06, 499.96it/s]
 37%|███▋      | 1911/5153 [00:03<00:06, 498.02it/s]
 38%|███▊      | 1961/5153 [00:03<00:06, 498.43it/s]
 39%|███▉      | 2011/5153 [00:04<00:06, 498.74it/s]
 40%|███▉      | 2061/5153 [00:04<00:06, 498.49it/s]
 41%|████      | 2118/5153 [00:04<00:05, 518.87it/s]
 42%|████▏     | 2179/5153 [00:04<00:05, 543.64it/s]
 43%|████▎     | 2239/5153 [00:04<00:05, 557.72it/s]
 45%|████▍     | 2299/5153 [00:04<00:05, 568.93it/s]
 46%|████▌     | 2359/5153 [00:04<00:04, 576.67it/s]
 47%|████▋     | 2419/5153 [00:04<00:04, 581.98it/s]
 48%|████▊     | 2478/5153 [00:04<00:04, 580.88it/s]
 49%|████▉     | 2538/5153 [00:04<00:04, 586.22it/s]
 50%|█████     | 2598/5153 [00:05<00:04, 588.55it/s]
 52%|█████▏    | 2658/5153 [00:05<00:04, 591.76it/s]
 53%|█████▎    | 2718/5153 [00:05<00:04, 591.51it/s]
 54%|█████▍    | 2778/5153 [00:05<00:04, 590.69it/s]
 55%|█████▌    | 2838/5153 [00:05<00:03, 591.51it/s]
 56%|█████▌    | 2898/5153 [00:05<00:03, 590.69it/s]
 57%|█████▋    | 2958/5153 [00:05<00:03, 592.08it/s]
 59%|█████▊    | 3018/5153 [00:05<00:03, 591.70it/s]
 60%|█████▉    | 3078/5153 [00:05<00:03, 589.41it/s]
 61%|██████    | 3138/5153 [00:05<00:03, 590.06it/s]
 62%|██████▏   | 3198/5153 [00:06<00:03, 591.73it/s]
 63%|██████▎   | 3258/5153 [00:06<00:03, 591.90it/s]
 64%|██████▍   | 3318/5153 [00:06<00:03, 592.73it/s]
 66%|██████▌   | 3378/5153 [00:06<00:02, 594.32it/s]
 67%|██████▋   | 3438/5153 [00:06<00:02, 589.09it/s]
 68%|██████▊   | 3497/5153 [00:06<00:02, 586.67it/s]
 69%|██████▉   | 3557/5153 [00:06<00:02, 589.27it/s]
 70%|███████   | 3617/5153 [00:06<00:02, 591.89it/s]
 71%|███████▏  | 3677/5153 [00:06<00:02, 592.95it/s]
 73%|███████▎  | 3737/5153 [00:06<00:02, 593.50it/s]
 74%|███████▎  | 3797/5153 [00:07<00:02, 592.08it/s]
 75%|███████▍  | 3857/5153 [00:07<00:02, 592.66it/s]
 76%|███████▌  | 3917/5153 [00:07<00:02, 593.40it/s]
 77%|███████▋  | 3977/5153 [00:07<00:01, 593.38it/s]
 78%|███████▊  | 4038/5153 [00:07<00:01, 595.67it/s]
 80%|███████▉  | 4098/5153 [00:07<00:01, 596.43it/s]
 81%|████████  | 4159/5153 [00:07<00:01, 597.80it/s]
 82%|████████▏ | 4219/5153 [00:07<00:01, 597.09it/s]
 83%|████████▎ | 4280/5153 [00:07<00:01, 598.09it/s]
 84%|████████▍ | 4340/5153 [00:07<00:01, 596.62it/s]
 85%|████████▌ | 4401/5153 [00:08<00:01, 598.04it/s]
 87%|████████▋ | 4461/5153 [00:08<00:01, 596.95it/s]
 88%|████████▊ | 4522/5153 [00:08<00:01, 598.53it/s]
 89%|████████▉ | 4583/5153 [00:08<00:00, 599.44it/s]
 90%|█████████ | 4644/5153 [00:08<00:00, 600.79it/s]
 91%|█████████▏| 4705/5153 [00:08<00:00, 600.70it/s]
 92%|█████████▏| 4766/5153 [00:08<00:00, 601.40it/s]
 94%|█████████▎| 4827/5153 [00:08<00:00, 602.13it/s]
 95%|█████████▍| 4888/5153 [00:08<00:00, 601.58it/s]
 96%|█████████▌| 4949/5153 [00:09<00:00, 601.89it/s]
 97%|█████████▋| 5010/5153 [00:09<00:00, 601.71it/s]
 98%|█████████▊| 5071/5153 [00:09<00:00, 599.86it/s]
+2026-09-16:23:36:34 INFO     [evaluator:589] Running loglikelihood requests
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+bootstrapping for stddev: perplexity
+
  0%|          | 0/100 [00:00<?, ?it/s]
  1%|          | 1/100 [00:04<07:13,  4.38s/it]
  4%|▍         | 4/100 [00:04<01:22,  1.17it/s]
  6%|▌         | 6/100 [00:04<00:54,  1.71it/s]
 64%|██████▍   | 64/100 [00:05<00:01, 29.54it/s]
+fatal: not a git repository (or any parent up to mount point /)
+Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
+2026-09-16:23:39:38 INFO     [loggers.evaluation_tracker:247] Saving results aggregated
+2026-09-16:23:39:38 INFO     [loggers.evaluation_tracker:119] Saving per-task samples to /workspace/loopq_data/curve_a4_s0/eval_final/lambada_openai/harness/ByteDance__Ouro-1.4B/*.jsonl
+vllm ({'pretrained': 'ByteDance/Ouro-1.4B', 'revision': 'e3b1e0993b1231a51d6069a870476dda4162c00a', 'dtype': 'bfloat16', 'tensor_parallel_size': 1, 'max_model_len': 4096, 'gpu_memory_utilization': 0.9, 'enforce_eager': True, 'max_num_seqs': 32, 'max_num_batched_tokens': 4096, 'enable_prefix_caching': False, 'seed': 0}), gen_kwargs: ({}), limit: None, num_fewshot: 0, batch_size: auto
+|    Tasks     |Version|Filter|n-shot|  Metric  |   |Value |   |Stderr|
+|--------------|------:|------|-----:|----------|---|-----:|---|-----:|
+|lambada_openai|      1|none  |     0|acc       |↑  |0.5605|±  |0.0069|
+|              |       |none  |     0|perplexity|↓  |8.3831|±  |0.2310|
+

loopq_quantization/results/loopq_w4a4_completed/lambada_openai/harness/ByteDance__Ouro-1.4B/results_2026-09-16T23-39-38.030831.json

@@ -0,0 +1,152 @@
+{
+  "results": {
+    "lambada_openai": {
+      "name": "lambada_openai",
+      "alias": "lambada_openai",
+      "sample_len": 5153,
+      "perplexity,none": 8.383131365672106,
+      "perplexity_stderr,none": 0.23103998744055027,
+      "acc,none": 0.5604502231709684,
+      "acc_stderr,none": 0.006914879684264954
+    }
+  },
+  "group_subtasks": {},
+  "configs": {
+    "lambada_openai": {
+      "task": "lambada_openai",
+      "dataset_path": "EleutherAI/lambada_openai",
+      "dataset_name": "default",
+      "test_split": "test",
+      "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
+      "doc_to_target": "{{' '+text.split(' ')[-1]}}",
+      "unsafe_code": false,
+      "description": "",
+      "target_delimiter": " ",
+      "fewshot_delimiter": "\n\n",
+      "fewshot_config": {
+        "sampler": "default",
+        "split": null,
+        "process_docs": null,
+        "fewshot_indices": null,
+        "samples": null,
+        "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
+        "doc_to_choice": null,
+        "doc_to_target": "{{' '+text.split(' ')[-1]}}",
+        "gen_prefix": null,
+        "fewshot_delimiter": "\n\n",
+        "target_delimiter": " "
+      },
+      "num_fewshot": 0,
+      "metric_list": [
+        {
+          "metric": "perplexity",
+          "aggregation": "perplexity",
+          "higher_is_better": false
+        },
+        {
+          "metric": "acc",
+          "aggregation": "mean",
+          "higher_is_better": true
+        }
+      ],
+      "output_type": "loglikelihood",
+      "repeats": 1,
+      "should_decontaminate": true,
+      "doc_to_decontamination_query": "{{text}}",
+      "metadata": {
+        "version": 1.0,
+        "pretrained": "ByteDance/Ouro-1.4B",
+        "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+        "trust_remote_code": true,
+        "dtype": "bfloat16",
+        "tensor_parallel_size": 1,
+        "max_model_len": 4096,
+        "gpu_memory_utilization": 0.9,
+        "enforce_eager": true,
+        "max_num_seqs": 32,
+        "max_num_batched_tokens": 4096,
+        "enable_prefix_caching": false,
+        "seed": 0,
+        "config_source": "/root/venv-LT/lib/python3.11/site-packages/lm_eval/tasks/lambada/lambada_openai.yaml"
+      }
+    }
+  },
+  "versions": {
+    "lambada_openai": 1.0
+  },
+  "n-shot": {
+    "lambada_openai": 0
+  },
+  "higher_is_better": {
+    "lambada_openai": {
+      "perplexity": false,
+      "acc": true
+    }
+  },
+  "n-samples": {
+    "lambada_openai": {
+      "original": 5153,
+      "effective": 5153
+    }
+  },
+  "config": {
+    "model": "vllm",
+    "model_args": {
+      "pretrained": "ByteDance/Ouro-1.4B",
+      "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+      "trust_remote_code": true,
+      "dtype": "bfloat16",
+      "tensor_parallel_size": 1,
+      "max_model_len": 4096,
+      "gpu_memory_utilization": 0.9,
+      "enforce_eager": true,
+      "max_num_seqs": 32,
+      "max_num_batched_tokens": 4096,
+      "enable_prefix_caching": false,
+      "seed": 0
+    },
+    "batch_size": "auto",
+    "batch_sizes": [],
+    "device": "cuda:0",
+    "use_cache": null,
+    "limit": null,
+    "bootstrap_iters": 100000,
+    "gen_kwargs": {},
+    "random_seed": 0,
+    "numpy_seed": 0,
+    "torch_seed": 0,
+    "fewshot_seed": 0
+  },
+  "git_hash": null,
+  "date": 1789601746.1443465,
+  "pretty_env_info": "PyTorch version: 2.8.0+cu128\nIs debug build: False\nCUDA used to build PyTorch: 12.8\nROCM used to build PyTorch: N/A\n\nOS: Ubuntu 22.04.5 LTS (x86_64)\nGCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0\nClang version: Could not collect\nCMake version: Could not collect\nLibc version: glibc-2.35\n\nPython version: 3.11.14 | packaged by conda-forge | (main, Oct 13 2025, 14:09:32) [GCC 14.3.0] (64-bit runtime)\nPython platform: Linux-6.8.0-124-generic-x86_64-with-glibc2.35\nIs CUDA available: True\nCUDA runtime version: Could not collect\nCUDA_MODULE_LOADING set to: LAZY\nGPU models and configuration: \nGPU 0: NVIDIA A40\nGPU 1: NVIDIA A40\n\nNvidia driver version: 580.159.04\ncuDNN version: Could not collect\nHIP runtime version: N/A\nMIOpen runtime version: N/A\nIs XNNPACK available: True\n\nCPU:\nArchitecture:                            x86_64\nCPU op-mode(s):                          32-bit, 64-bit\nAddress sizes:                           52 bits physical, 57 bits virtual\nByte Order:                              Little Endian\nCPU(s):                                  96\nOn-line CPU(s) list:                     0-95\nVendor ID:                               GenuineIntel\nModel name:                              Intel(R) Xeon(R) Gold 6342 CPU @ 2.80GHz\nCPU family:                              6\nModel:                                   106\nThread(s) per core:                      2\nCore(s) per socket:                      24\nSocket(s):                               2\nStepping:                                6\nCPU max MHz:                             3500.0000\nCPU min MHz:                             800.0000\nBogoMIPS:                                5600.00\nFlags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect wbnoinvd dtherm ida arat pln pts vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities\nVirtualization:                          VT-x\nL1d cache:                               2.3 MiB (48 instances)\nL1i cache:                               1.5 MiB (48 instances)\nL2 cache:                                60 MiB (48 instances)\nL3 cache:                                72 MiB (2 instances)\nNUMA node(s):                            2\nNUMA node0 CPU(s):                       0-23,48-71\nNUMA node1 CPU(s):                       24-47,72-95\nVulnerability Gather data sampling:      Mitigation; Microcode\nVulnerability Indirect target selection: Mitigation; Aligned branch/return thunks\nVulnerability Itlb multihit:             Not affected\nVulnerability L1tf:                      Not affected\nVulnerability Mds:                       Not affected\nVulnerability Meltdown:                  Not affected\nVulnerability Mmio stale data:           Mitigation; Clear CPU buffers; SMT vulnerable\nVulnerability Reg file data sampling:    Not affected\nVulnerability Retbleed:                  Not affected\nVulnerability Spec rstack overflow:      Not affected\nVulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl\nVulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization\nVulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop\nVulnerability Srbds:                     Not affected\nVulnerability Tsa:                       Not affected\nVulnerability Tsx async abort:           Not affected\nVulnerability Vmscape:                   Not affected\n\nVersions of relevant libraries:\n[pip3] numpy==2.2.6\n[pip3] nvidia-cublas-cu12==12.8.4.1\n[pip3] nvidia-cuda-cupti-cu12==12.8.90\n[pip3] nvidia-cuda-nvrtc-cu12==12.8.93\n[pip3] nvidia-cuda-runtime-cu12==12.8.90\n[pip3] nvidia-cudnn-cu12==9.10.2.21\n[pip3] nvidia-cufft-cu12==11.3.3.83\n[pip3] nvidia-curand-cu12==10.3.9.90\n[pip3] nvidia-cusolver-cu12==11.7.3.90\n[pip3] nvidia-cusparse-cu12==12.5.8.93\n[pip3] nvidia-cusparselt-cu12==0.7.1\n[pip3] nvidia-nccl-cu12==2.27.3\n[pip3] nvidia-nvjitlink-cu12==12.8.93\n[pip3] nvidia-nvtx-cu12==12.8.90\n[pip3] torch==2.8.0\n[pip3] torchaudio==2.8.0\n[pip3] torchvision==0.23.0\n[pip3] triton==3.4.0\n[conda] Could not collect",
+  "transformers_version": "4.56.2",
+  "lm_eval_version": "0.4.13.dev0",
+  "upper_git_hash": null,
+  "tokenizer_pad_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "tokenizer_eos_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "tokenizer_bos_token": [
+    "<|endoftext|>",
+    "0"
+  ],
+  "eot_token_id": 0,
+  "max_length": 4096,
+  "task_hashes": {
+    "lambada_openai": "35c949e9b587d74c4161028a77e63b13e17f1c1441e6441d236ee778ce6b0959"
+  },
+  "model_source": "vllm",
+  "model_name": "ByteDance/Ouro-1.4B",
+  "model_name_sanitized": "ByteDance__Ouro-1.4B",
+  "system_instruction": null,
+  "system_instruction_sha": null,
+  "fewshot_as_multiturn": null,
+  "chat_template": null,
+  "chat_template_sha": null,
+  "total_evaluation_time_seconds": "237.92010679794475"
+}

loopq_quantization/results/loopq_w4a4_completed/lambada_openai/installed_task_snapshot/README.md

@@ -0,0 +1,39 @@
+# LAMBADA
+
+### Paper
+Title: `The LAMBADA dataset: Word prediction requiring a broad discourse context`
+
+Abstract: https://arxiv.org/pdf/1606.06031.pdf
+
+LAMBADA is a dataset to evaluate the capabilities of computational models for text
+understanding by means of a word prediction task. LAMBADA is a collection of narrative
+passages sharing the characteristic that human subjects are able to guess their last
+word if they are exposed to the whole passage, but not if they only see the last
+sentence preceding the target word. To succeed on LAMBADA, computational models
+cannot simply rely on local context, but must be able to keep track of information
+in the broader discourse.
+
+Homepage: https://zenodo.org/record/2630551#.X4Xzn5NKjUI
+
+### Groups and Tasks
+
+#### Groups
+
+- `lambada`
+
+#### Tasks
+
+- `lambada_openai`
+- `lambada_standard`
+
+
+### Citation
+
+@misc{
+    author={Paperno, Denis and Kruszewski, Germán and Lazaridou, Angeliki and Pham, Quan Ngoc and Bernardi, Raffaella and Pezzelle, Sandro and Baroni, Marco and Boleda, Gemma and Fernández, Raquel},
+    title={The LAMBADA dataset},
+    DOI={10.5281/zenodo.2630551},
+    publisher={Zenodo},
+    year={2016},
+    month={Aug}
+}

loopq_quantization/results/loopq_w4a4_completed/lambada_openai/installed_task_snapshot/lambada_openai.yaml

@@ -0,0 +1,20 @@
+tag:
+  - lambada
+task: lambada_openai
+dataset_path: EleutherAI/lambada_openai
+dataset_name: default
+output_type: loglikelihood
+test_split: test
+doc_to_text: "{{text.split(' ')[:-1]|join(' ')}}"
+doc_to_target: "{{' '+text.split(' ')[-1]}}"
+should_decontaminate: true
+doc_to_decontamination_query: "{{text}}"
+metric_list:
+  - metric: perplexity
+    aggregation: perplexity
+    higher_is_better: false
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true
+metadata:
+  version: 1.0

loopq_quantization/results/loopq_w4a4_completed/lambada_openai/installed_task_snapshot/lambada_standard.yaml

@@ -0,0 +1,21 @@
+tag:
+  - lambada
+task: lambada_standard
+dataset_path: cimec/lambada
+dataset_name: null
+output_type: loglikelihood
+validation_split: validation
+test_split: test
+doc_to_text: "{{text.split(' ')[:-1]|join(' ')}}"
+doc_to_target: "{{' '+text.split(' ')[-1]}}"
+should_decontaminate: true
+doc_to_decontamination_query: "{{text}}"
+metric_list:
+  - metric: perplexity
+    aggregation: perplexity
+    higher_is_better: false
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true
+metadata:
+  version: 1.0

loopq_quantization/results/loopq_w4a4_completed/lambada_openai/run_metadata.json

@@ -0,0 +1,721 @@
+{
+  "status": "completed",
+  "task": "lambada_openai",
+  "configuration": "loopq_w4a4",
+  "contract": {
+    "mode": "loopq",
+    "activation_bits": 4,
+    "artifact": "/workspace/loopq_data/curve_a4_s0/eval_final/artifact.pt",
+    "artifact_sha256": "7cb498143aeb807e5cb896f195066fa07e38a353406a10de36ffaf46b4fcabdb",
+    "ablation": null,
+    "parity_evidence": {
+      "path": "/workspace/loopq_data/curve_a4_s0/eval_final/parity_screen.json",
+      "sha256": "0492028e64eaf16242d69d32bec9a7783e3637875ddaae3b986889c86a1b04ed",
+      "acceptance": {
+        "passed": false,
+        "reasons": [
+          "four-loop hidden-state parity failed",
+          "logprob error exceeds tolerance",
+          "top1 or top-1000 comparison incomplete"
+        ],
+        "max_logprob_error": 1.25,
+        "max_hidden_error": 0.06,
+        "tie_threshold": 0.25,
+        "rows_excused_as_reference_tie": 0,
+        "scope": "six_or_more_single_sequence_prefills; decode/batch/long_context still require A40 validation",
+        "advisory": true
+      }
+    }
+  },
+  "model_args": {
+    "pretrained": "ByteDance/Ouro-1.4B",
+    "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+    "trust_remote_code": true,
+    "dtype": "bfloat16",
+    "tensor_parallel_size": 1,
+    "max_model_len": 4096,
+    "gpu_memory_utilization": 0.9,
+    "enforce_eager": true,
+    "max_num_seqs": 32,
+    "max_num_batched_tokens": 4096,
+    "enable_prefix_caching": false,
+    "seed": 0
+  },
+  "fewshot": 0,
+  "seed": 0,
+  "smoke_limit": null,
+  "study_evaluation": false,
+  "protocol_note": "LoopQ W4A4, calibration run to completion (1500 final optimization steps) with singular values held in log space so they cannot reach zero. The paper does not specify how the Kronecker transforms are kept invertible. Deployment prefill-parity gate is unreachable at W4A4 by any implementation and is attached as advisory. Explicit local protocol: zero-shot except MMLU 5-shot; plain likelihood scoring, no chat template; installed pinned harness prompts.",
+  "metric_note": "LAMBADA last-word acc and perplexity",
+  "exact_paper_protocol_reproduction": false,
+  "disclosure": "Single explicit-seed run; paper reports ten-run accuracy averages. Dataset/harness choices must be compared before claiming reproduction.",
+  "command": [
+    "/root/venv-LT/bin/python",
+    "/workspace/LT_fix/scripts/run_lm_eval_timed.py",
+    "--model",
+    "vllm",
+    "--model_args",
+    "{\"pretrained\": \"ByteDance/Ouro-1.4B\", \"revision\": \"e3b1e0993b1231a51d6069a870476dda4162c00a\", \"trust_remote_code\": true, \"dtype\": \"bfloat16\", \"tensor_parallel_size\": 1, \"max_model_len\": 4096, \"gpu_memory_utilization\": 0.9, \"enforce_eager\": true, \"max_num_seqs\": 32, \"max_num_batched_tokens\": 4096, \"enable_prefix_caching\": false, \"seed\": 0}",
+    "--tasks",
+    "lambada_openai",
+    "--num_fewshot",
+    "0",
+    "--seed",
+    "0,0,0,0",
+    "--batch_size",
+    "auto",
+    "--output_path",
+    "/workspace/loopq_data/curve_a4_s0/eval_final/lambada_openai/harness",
+    "--log_samples"
+  ],
+  "environment": {
+    "torch": "2.8.0",
+    "vllm": "0.10.2",
+    "transformers": "4.56.2",
+    "lm_eval": "0.4.13.dev0",
+    "datasets": "5.0.1"
+  },
+  "returncode": 0,
+  "wall_clock_seconds": 246.9097606409341,
+  "source_provenance": {
+    "snapshot": "/workspace/loopq_data/curve_a4_s0/eval_final/lambada_openai/source_snapshot",
+    "sha256_at_launch": {
+      "loopQ/adapters/__init__.py": "7f28e14f7778bafb750404e0e54fb05fe17b61997e06584ca55ee44da3694ba7",
+      "loopQ/adapters/huginn.py": "fc22beafc461a7a0f85e6a7e3291a60749d5a11ea56cceabc448239dd3be9afe",
+      "loopQ/adapters/ouro.py": "e571df0cbbe853de8d1c593f9ed6c032e9099008363ee57e1f280b770ae49c14",
+      "loopQ/configs/flatquant_reference.yaml": "35f507d7389c3f04a7c1f86f7ae367630c846ec730661d420639f5f19b0c4074",
+      "loopQ/configs/huginn_mapping.yaml": "5db252e7cacb42169e94492e02509a3114ebca4747b32590110ba58cbbaea124",
+      "loopQ/configs/loopq_compatibility_decisions.yaml": "3d1b6d508db78b53f2c6e405da3b8ea6a7893303bd81bc2a90e0b3567367fd35",
+      "loopQ/configs/ouro_calibration.yaml": "fc832996fcdda085290565b657c6fa1824dc0b1646a66d109fe38b22f0a8eac5",
+      "loopQ/configs/ouro_mapping.yaml": "87ca4718e200539412f366fbad52a1a46932ad2cc26e0f08585830d04f50b9ad",
+      "loopQ/configs/paper_ouro_reference_metrics.json": "ecc4b3e028b1191a458b446582895d4f6871922a397f15375a34e396bf70ad89",
+      "loopQ/configs/paper_specification.yaml": "c812bb40398d9183f522c07e32dabf3fcfdd43e4d52790b9af296fb11d5abed1",
+      "loopQ/configs/parity_prompts.json": "b56e2e4048083459997c2864e3e6320fc68e1625676ed09f207a81bafee9b2e0",
+      "loopQ/configs/reproduction.json": "5204e346d98f96e46b0f94814e2f291198887ace85a7c827932f002560c350c3",
+      "loopQ/loopq/__init__.py": "00cf79e9673051148513cfc25c52c59ef72118d288e2ec640a77eaf8a69d7213",
+      "loopQ/loopq/accounting.py": "ab24ef8b10a0bc85e03827e882f92e1d7bb3e33fbd146f7f9561651842a03fb0",
+      "loopQ/loopq/calibrate.py": "5018098ddf0c44a682efef85b9f82ab2e4f4f911efa48982083d6c3a0db98594",
+      "loopQ/loopq/cta.py": "e1cf8cfec967157a90d9f54e956b221b67aef9ed742968605d2d4abb1f453b8d",
+      "loopQ/loopq/evaluation_provenance.py": "87e0d6f329dae65759893063cf508682902ae4cf5772aec28ed1692a08da02cf",
+      "loopQ/loopq/experiment_contract.py": "a0f4487092a72eb5fe808880d09a6fafc9b86e5fc4a2e2b3b9e7d5177d27a620",
+      "loopQ/loopq/experiments.py": "252644613f076890e5f2b849c4f95570e33a9cfa69569df90b082c26788f5a5b",
+      "loopQ/loopq/export.py": "3ca272d220bbc93d2e7c70b06df2cef93184e48bc695c00b97e834991469ef57",
+      "loopQ/loopq/gpu_admission.py": "feb940b5cc2355f1881d031f9b22226d7692f70f792c1b1233aeec5570535c76",
+      "loopQ/loopq/huginn_calibration.py": "7a95635b9d9777c9b0f04f51566f9ed668154ae9be231e19f3b7ac18acef98ca",
+      "loopQ/loopq/journal.py": "f9bf8342a004959bbdb25b3557577e5271e7335f53107d7a007ad512fa55bbcf",
+      "loopQ/loopq/las.py": "a179c62e57fccafc21dabb13a68832f8bc8c9d6b0a8da1a2c2348bfb09c47a3c",
+      "loopQ/loopq/numerical_contract.py": "913685841222a9f7ff503ca9b8962cd976f39bda38e2a20f7d3e6b0c09798bc5",
+      "loopQ/loopq/objective.py": "d60f1d6ce18e004f89cb91f680ed6dd4b28083dda8125745a1bd694389c40e50",
+      "loopQ/loopq/ouro_calibration.py": "db88d48b9646c421e4fd354e8de61a7784bdedcbac0b47e1c8de01aa8c2395de",
+      "loopQ/loopq/packed_bundle.py": "5fe29dee51fe28849696de298d3bc8f7036b78c44ec2617747ada370e135287f",
+      "loopQ/loopq/packed_weight.py": "8bbd5a3dc0c8992e4dd30ce962deb0e0745604b1e189a508e1f70aee1e1acdbc",
+      "loopQ/loopq/packed_worker.py": "239febb2ed17d9d88cb068d2edac87f13c2f5c579995a842cdfb7ab6b1e536aa",
+      "loopQ/loopq/parity_worker.py": "b6854262f3b44e4970214aa3e8ccc5a8ad1fa032a9ae5314a15514e132540502",
+      "loopQ/loopq/paths.py": "5a9478359f3a4b252d5af599c9c7f68650d10b995f2e9314210ac6e670a364d0",
+      "loopQ/loopq/provenance.py": "387f82f3efe26ea478476d7087db441a2552f345de35ce2afc5723eb4d21abd6",
+      "loopQ/loopq/quantization.py": "78a3016ad8b2d53ae807bb18b71da8e7d8ce6df8d0881331a80776a247c1142e",
+      "loopQ/loopq/readiness.py": "837a59bc6ac641ea2140876fd3ae34afc30173b91791852ba55dafd4bdd9cdd8",
+      "loopQ/loopq/saved_tensors.py": "ee158cc09d8116c09922b940d3219d7d22d3f271b9c11e8b623b393cecb2af03",
+      "loopQ/loopq/sharing_gap.py": "aa6e1bdb0f365fdde0e9ed4674a08b0f2c0df0a0693dd28dc5c094f8f86118a5",
+      "loopQ/loopq/trajectory_diagnostics.py": "de78ef945d1fdcd036460ad5947864f73a5d63a5f88a1a6c86691821b2dddcac",
+      "loopQ/loopq/transforms.py": "d754f21df55fde2548872d25afa9a4fe75082fdc937a3ba65b474a0673be12a6",
+      "loopQ/scripts/benchmark_calibration_memory.py": "afdaf39cde1e839924baf1cfd0a4d83df5ee0af40d6e7659cb6f3645e3ebb408",
+      "loopQ/scripts/calibrate_huginn.py": "ba44cc894b5a9987cb2f3a1382053d907f205a2d7608b44b7714fc410c142756",
+      "loopQ/scripts/calibrate_ouro.py": "5c81cb8ca6a92a6da4fe1b7e5d7615830e359d93a4eedf013fbb32d5f624ec9e",
+      "loopQ/scripts/diagnose_ouro_quality.py": "962f5ac918bea5cb029c78f41b96b886013e8233d5bf1d67ec87d8b488b9977c",
+      "loopQ/scripts/export_huginn_artifact.py": "20dbf608320f85daddbc843b0ac1196234c20822e2418f42b66327ddcd8186f7",
+      "loopQ/scripts/export_ouro_artifact.py": "1fe2edfa0fc09cf896a1fab149067160829744d3f7fd03d2c521b6ba38d19a12",
+      "loopQ/scripts/export_packed_ouro_weights.py": "a49b8646639682012340f908ec86efbe12a369343c984963c2314991c31a4ead",
+      "loopQ/scripts/freeze_lambada_inputs.py": "b15138ae82a52461f31cbb669d18f878fd98f9eab93ac7db3efa87f594afb97b",
+      "loopQ/scripts/inspect_calibration.py": "6514bb2fb9fda6ef6b5335035dae910cb4bf7ec36dde5a844c7cf38e0fd07f7a",
+      "loopQ/scripts/measure_ouro_trajectory.py": "06bf7ea56bb6626b5947e8f43d2b7ddc60ad942d3997ac1c2b7795db4672c081",
+      "loopQ/scripts/report_accounting.py": "3564bf5cb9e4f8133257cb415ec679f007090f0c7afe85fe1bca4b469b5ba5e6",
+      "loopQ/scripts/reproduction.py": "6a28f7a4cb1f4a9f4171fcf466b36b103761bc38bde63e3afefe8a25aca86fb1",
+      "loopQ/scripts/run_gsm8k.py": "11b220ad74277b5da6becbf18eaaeb4725d6ecb07330b1b15cd4dc4c29eeefdb",
+      "loopQ/scripts/run_paper_benchmark.py": "021ee1451abff633bc0b77f04c13e1cd9dae7bebfc86ebe962890015a74a6bf8",
+      "loopQ/scripts/smoke_huginn_vllm.py": "9311b038ce325a3f050a852addb6a9c550a09ace19f29ee3370439be9047e29b",
+      "loopQ/scripts/smoke_ouro_vllm.py": "39311a193900e7d1db491e29fcdd83ee88ad1bdda602025580f61bd5ae757203",
+      "loopQ/scripts/verify_runtime_parity.py": "1e0e1648fe5673abaca26d98efffac99cc3b2ede7ecb59423df04147e537e4e0",
+      "loopQ/tests/conftest.py": "56cfc62ddfe4ccab80288f60c8f3948b6d8b8d8f1d2e37c50f9c9577c4679235",
+      "loopQ/tests/test_alignment_contracts.py": "615d2e77c187411d55892fa875835ea280a9b6ab775835ed30459482c298f066",
+      "loopQ/tests/test_calibrate.py": "d9bddf96ebcd0c5bdd0dd866f17e045d672abd73360e5e0771775edc4e67c8c8",
+      "loopQ/tests/test_cta.py": "afe35581d60dcd0b5ff1566ffe3a8dc332003cd031d9c9e788cff1c2f62f2307",
+      "loopQ/tests/test_evaluation_provenance.py": "62765331e2e5b0e400c51b0d7caba5cc70f3060ac60fdfa53e89fc0eea293933",
+      "loopQ/tests/test_experiment_setup.py": "840680d43db300613ccb55409ce9bd142f2eb82176d697813194648de2c57484",
+      "loopQ/tests/test_export.py": "7bf542a1e606c7be517c2485dcad3b8e8ec773f53e7374656564ce33a78a6b8a",
+      "loopQ/tests/test_gpu_admission.py": "5f0154e0e7d80e1de684b51219b3ee1b52d81481d062de13a563eab1db17fce6",
+      "loopQ/tests/test_gsm8k_driver.py": "054d0284905dc8479ceb6c28e2a418d94a692762d45573902a8bb0278abd2821",
+      "loopQ/tests/test_huginn_adapter.py": "23f7b5d2fea8c9526553e979a5c44e466fe74d664cf6f0a05b8c66434dc48940",
+      "loopQ/tests/test_huginn_calibration.py": "bbdc7f0e10f23042f6d52acf14456d039b030d945fb4d3a8c72d9d6063242a93",
+      "loopQ/tests/test_huginn_export.py": "896eee3b771ee2a2b0dbdbd9aa07eb9918718c7dc9e69e0169b3d4ab22e50229",
+      "loopQ/tests/test_huginn_vllm_bridge.py": "c483f52999ca955b4c8b17de252e957bf7a878bfa7e5fa18330c7fed3f249c9b",
+      "loopQ/tests/test_journal.py": "b3331cbff1add620c8e1d098578a9d79f81e2c1d28a2e0c0b4a7a7a0c7ba1589",
+      "loopQ/tests/test_las.py": "88139ae1f23a53d07eb3497b88fe3f22980d6b8c001232fc3b629a57663935e4",
+      "loopQ/tests/test_objective.py": "fb206b4844e1f911e6665ccd5d99e643171810a30da50a80be64d227019e9b88",
+      "loopQ/tests/test_ouro_adapter.py": "9a52daa93ea65a75fc9edfd154ab27058c6a2567d113cab25e6d5a9e6592d986",
+      "loopQ/tests/test_ouro_calibration.py": "e9099d1cddec60482c811df656e7aa8731c2a1cd98ae274efde249636d0cec50",
+      "loopQ/tests/test_packed_bundle.py": "cfffba2fd0d640e2e5def1d6375b24a87b4d33bb57a7c45fb5d08145ae46bd4d",
+      "loopQ/tests/test_packed_weight.py": "387fdb51e3bce2cbd7c0f94dcec5fb1ecd8eedd55c79b6811975503625be884f",
+      "loopQ/tests/test_packed_worker.py": "41206a263d72bdc2e7ba65c8ecf7e6982f3b31b62c9602e554798212f8373a37",
+      "loopQ/tests/test_paper_benchmark.py": "70eaa2326bf0b8da12736a1577421a42af446aacd48fc21be0e9ead2deab52ed",
+      "loopQ/tests/test_paper_math_reference.py": "a46176f310f90d75bef72674620a2425de13a20bb65f84d33686dc82ff4043fb",
+      "loopQ/tests/test_quantization.py": "a0368909de90a7b5d64d0cb82296ab438306f4e337c50ed1a103616fdd3f9af4",
+      "loopQ/tests/test_saved_tensors.py": "522eae339ca2d2b01506f653b18d6417bc04b4315d0d1afa90b78820d55c7742",
+      "loopQ/tests/test_sharing_gap.py": "aacbacadaee3850d1d823cf8122cbc2ac80f2af3066254a7b5b60d46c0ded1ac",
+      "loopQ/tests/test_trajectory_diagnostics.py": "341d8650a0c3ae59c12584baaccc6fec454bb76f988e533385f5953d3f83bf3e",
+      "loopQ/tests/test_transforms.py": "5e9fc5b387fb8f40f3b090864a0c3751529a869bef176058bac149dd4401bfca",
+      "loopQ/tests/test_vllm_bridge.py": "ccab13d1a9c3bd18279e46b4268cccba84e766153282e9d78cd58e093fe50647",
+      "src/loop_transformer_quant/__init__.py": "a9dceaa469212264a6ee7b1a615de2c5cd6e6195a007c852fc7e1e85857c172f",
+      "src/loop_transformer_quant/adapters/__init__.py": "6d112e61f7a6dc6093f11278d03a5351e8dd7adcb55d34dbb9e6810f86dbee74",
+      "src/loop_transformer_quant/adapters/vllm.py": "9f8df73df24a2b2297def5f0b994b8508fd06dc1a16184af64220e9016b813fd",
+      "src/loop_transformer_quant/cache_codec.py": "cf22a2cda1546e2bacb1e7e6ce9d8bdf8db952d758dd5cf81fdc1e784154185e",
+      "src/loop_transformer_quant/dynamic_lse.py": "ca0ce130dd1f838ab34c96ba8fe23f308ba09975801f587e4d5fa5723ddd93f5",
+      "src/loop_transformer_quant/gsm8k_calibration.py": "f90b5b89e595ef86ca6d03ae4f77bfa62b123aee3b9efcacfa468a87c72d64fc",
+      "src/loop_transformer_quant/huginn_anchor_mechanisms.py": "bb59f9f3d90ff0bf3dada6cd7d48ddf26169796a0553a4c2a39fd520c2222b8a",
+      "src/loop_transformer_quant/int2_experiments.py": "7be8a00a688a57ee7aa0e8fbfe6e8adf15277dcacabd95e608b5707d9010abba",
+      "src/loop_transformer_quant/int2_runtime.py": "2eb7f992fe80c63874efb440f8ed3b1c84374d16118872532d9ada3cde3a0b1e",
+      "src/loop_transformer_quant/int2_streaming.py": "e16f5ab4abd71af8e3ea5ec3cc3164a5956bacbe87cd81f1b7c2af551bb2bd75",
+      "src/loop_transformer_quant/kv_codecs.py": "118c98e0e32d899d7e6458d8c626738912af9ce87d761789a0d7d2a51e8d24c6",
+      "src/loop_transformer_quant/layer_channel_heatmaps.py": "04f09e794e6f87d75410a2ff55bbd60f047bdaf5355dcb4abca085b0a0fb318f",
+      "src/loop_transformer_quant/logical_storage.py": "01858edbf3c5406af60c747375c0fbee6b406fc867a0ea1741525caeb90245f3",
+      "src/loop_transformer_quant/native_activation.py": "4f82f90e80574e92e2dd97128a206983f32e0bb424286fad33ceee91c19dd366",
+      "src/loop_transformer_quant/native_kv/__init__.py": "acfa90ad9c90018df7f68d2efecfd8f9dbb4ca90344279ba3aa307e6829a5b39",
+      "src/loop_transformer_quant/native_kv/cache.py": "573b46adf72ad17415ad2e1a8af977cefa63f3958698fee7ecdc50b2a8a8ff2a",
+      "src/loop_transformer_quant/native_kv/calibration.py": "6b5123c4933e147f44c8a47c339311a034f842957155894f81eeea0f4be0e674",
+      "src/loop_transformer_quant/native_kv/channel_lse_certified.py": "43d0d6dcce1de9fff4c99979336a0861c67f7bf6d7ba786509201ca03e8cd0cf",
+      "src/loop_transformer_quant/native_kv/channel_lse_cost_probe.py": "8a816598f39847d32318383e456dab2e575e04047ed89434fcc830428ece23f8",
+      "src/loop_transformer_quant/native_kv/channel_lse_fastproducts.py": "c7b4b4927725c7388342aee92abf014304e81714c27050bce4709725a0a3eb19",
+      "src/loop_transformer_quant/native_kv/channel_lse_guarded.py": "04f2893e42049ed01b7c0c4a4076f9ae5ad786562fc81311e2ab6ffbdca0fefc",
+      "src/loop_transformer_quant/native_kv/channel_lse_kahan.py": "036991681f5e6667f6684c0b794eefdfdcbdffa2b096095413544a848c96ba00",
+      "src/loop_transformer_quant/native_kv/channel_lse_mixedstats.py": "3acb22432e21870c99d83f12121cec2882bff1ad9b14ebc27600699b68ea9cc5",
+      "src/loop_transformer_quant/native_kv/channel_lse_native.py": "fe7d3082aba93d483983e091053d722860b51b478e138f92a27114c5c6ee7a11",
+      "src/loop_transformer_quant/native_kv/channel_lse_pagemixed.py": "2b83812cf5133e4a88780d4a89fa7179ec1ea581f68b4f356551d896e162a981",
+      "src/loop_transformer_quant/native_kv/channel_lse_pagetiled.py": "5f38be7492b93176cbe998d3b33345056a6db90c40275261b5c65ce3631cf104",
+      "src/loop_transformer_quant/native_kv/channel_lse_reference.py": "b09e33d551bbe31a6fd56469a207f8a6c9173eb915761c3c263c2574ee31567e",
+      "src/loop_transformer_quant/native_kv/channel_lse_solve32.py": "eeafd3685da864208cffefef9d5bd3575a8a2b873baf7f38e79dcbcf3fbd64f9",
+      "src/loop_transformer_quant/native_kv/channel_lse_split_solve.py": "d847cccf05382f38c337c094654e892c1544e37dff1fbb89967e349220756ccf",
+      "src/loop_transformer_quant/native_kv/channel_packed.py": "083f30a6446e125079df2e17811d06d004ee289953b999304f95e989acd86513",
+      "src/loop_transformer_quant/native_kv/channel_paged_fused.py": "a7fc2ab879859804913838b38c7888db589420f2fcdf5303165ff6a72c4d6f57",
+      "src/loop_transformer_quant/native_kv/channel_paged_grouped.py": "e52cf976ee34bd680af8107b18e97e50b7272deac2b3925e5c12f1d248171ca7",
+      "src/loop_transformer_quant/native_kv/channel_paged_meta.py": "179c3df4189e034f5c5b233f23e76bc0a49d12bab990480525d5f033ecc2ea6e",
+      "src/loop_transformer_quant/native_kv/channel_paged_recompute.py": "0ca703afd01d66337c949af2b3d73513bf801cfecb41f1c7851120e7e6641f63",
+      "src/loop_transformer_quant/native_kv/channel_prefix_scan.py": "3ef7845af24a20f515b2f04461c97fb2eed3d9d9200cc60d77bca4b7b34b0c1b",
+      "src/loop_transformer_quant/native_kv/channel_split_write.py": "8b197f54626758564c220aba8908de298ae2f4929a7131abbdf2d9fdea192421",
+      "src/loop_transformer_quant/native_kv/channel_token_fused.py": "eb455108f066f1ad90821a9484efc2ac4967f1b72d5ac83773054ff36ab16f36",
+      "src/loop_transformer_quant/native_kv/channel_token_meta.py": "443f3332b6e31cba118e880f71fed429c007f6d53543625d273b66eef4fe90a8",
+      "src/loop_transformer_quant/native_kv/comparison.py": "d318411d27db8f903ca1f250db8196d3b490e7892f9ee49d647dc3f304f88b00",
+      "src/loop_transformer_quant/native_kv/dynamic_paged.py": "c34a21557bd218278b6a08a8cdaa0b0a30a2b29a1ff02db02579f9cb7c73556b",
+      "src/loop_transformer_quant/native_kv/dynamic_paged_fused.py": "3c0f5c620c6bcb938bea32a94aa832c70fb356716088dfda207f47004727e393",
+      "src/loop_transformer_quant/native_kv/dynamic_scalar.py": "69e954ac3100cddb7464ec3e6caf9c3d3927fc96b99c472868057aa821a69867",
+      "src/loop_transformer_quant/native_kv/fa_tile.py": "5836fc96d54b5892fa911f0d0d48f81c9a91d71690c0455be639f31958bd77d0",
+      "src/loop_transformer_quant/native_kv/kernels.py": "d5537c45945724076984af5ff68c9e7b859528bf5c6ccc437c3885fc96f002f0",
+      "src/loop_transformer_quant/native_kv/ouro_comparison.py": "bf232a254102672863fe233e627a4e0059286a7f55e6362e397b22ce83178e77",
+      "src/loop_transformer_quant/native_kv/request_slots.py": "662be79c9ca67d8dace1c843a4ccb1b9325cb76b0b97c9c1fcdc677e44e86e25",
+      "src/loop_transformer_quant/native_kv/static_paged_fused.py": "19dfc2d58beac8b5318a97ca8fa2b5ac3badc4f33c0ea013d70e9db4d33d8466",
+      "src/loop_transformer_quant/native_kv/xqa_loader.py": "4de9c6c9f3a68d4fc422d68d3e55868dc756686299f901127acf284148472b0e",
+      "src/loop_transformer_quant/nvfp4.py": "68f0623c31f9667b295d09b2ecf4832aee84dbf93ab3de7348579189ca44a7d1",
+      "src/loop_transformer_quant/packed_affine_kv.py": "3c249789ca964aa1947262e9177fc3f66401d67889d1396a511cc489ea89e682",
+      "src/loop_transformer_quant/packed_rotated_hnd.py": "844421a98211c9a4bc1867c79f19c7d5583e82abe606b941276260c9a9e9c662",
+      "src/loop_transformer_quant/packed_rotated_hnd_safe.py": "9409fd0cca643ed8b3c87de99440e66eb7d3e9e8d6fdce8c76553806c3c29323",
+      "src/loop_transformer_quant/packed_rotated_kv.py": "195427fea9b110b693fbb54a1cd809482bdec7c75417055ec9ad14c8ed182ecb",
+      "src/loop_transformer_quant/packed_rotated_table1.py": "9b246daf90bf8d005482d413fd4f72422945d8ac001e98c61075365a6e1e8885",
+      "src/loop_transformer_quant/paths.py": "1f87fce7344da5916fd4a8b88dec2f5d7b45bf4b4cab129b72e9bf2280595d9c",
+      "src/loop_transformer_quant/pre_sinq.py": "e7926059546171473b26d1fdcfb2378884fd95aefe573d93119acf981e13240e",
+      "src/loop_transformer_quant/quant_policy.py": "bc55de3e7d27d7825cb2dbf443cd90ab3fe0452856a38864ecdb93c2e0fe3e38",
+      "src/loop_transformer_quant/rotated_attention_adaptive.py": "6ca61bf1207abe12e40de81900348df49d1832a69d1d2b9b92fee81daceedc34",
+      "src/loop_transformer_quant/rotated_attention_adaptive_safe.py": "5c93b4f091f723a1631b210bfb0589501c191d87adc57eacc8e90c98ae3b3f2c",
+      "src/loop_transformer_quant/rotated_attention_combined.py": "24d0fea4c099321594fe0d80c82093e934fd231137559d1a9d90535470db0458",
+      "src/loop_transformer_quant/rotated_attention_combined2.py": "207860a760e4a2bb2491d233247fabe965c84796ad7eb1bdab83df9a64d3d77c",
+      "src/loop_transformer_quant/rotated_attention_combined4.py": "df49a846ac8059bbdd9e5441fa91d366005063d0946c1c1fafff929998b5ca06",
+      "src/loop_transformer_quant/rotated_attention_conservative.py": "6318dbf853294db80b397985b0f3b92fce7d97791c752a0fb9df949095b4a294",
+      "src/loop_transformer_quant/rotated_attention_fast.py": "2e884106998c22e14fe822d3267478da54521ebfb93ab76e174daec23913ba9f",
+      "src/loop_transformer_quant/rotated_attention_group_all.py": "d6973ecff1f66eed99f9955b102f6ba5d51d240e249f724d6667028998da0347",
+      "src/loop_transformer_quant/rotated_attention_group_codes.py": "be0723b591b7c95436057c3ea7616c4178e516c8dc80495b698375196239e921",
+      "src/loop_transformer_quant/rotated_attention_group_codes_v2.py": "327119e221287a8221d178290aba72c4d428ff1d6664347374f04d287146da83",
+      "src/loop_transformer_quant/rotated_attention_group_load.py": "7cdda56e000afe900d7b6d3a942f112c4ab986a1a70476275531dc3238696063",
+      "src/loop_transformer_quant/rotated_attention_group_pipeline.py": "b2139be31958bb37fdf0afbfdfdbcde8ab91245208782ba16631ea47b369f3f3",
+      "src/loop_transformer_quant/rotated_attention_group_rotunroll.py": "f4bdc70394069dc078649b6ad5310ccbce2e5d9dff048f403fcebba687ce6fb5",
+      "src/loop_transformer_quant/rotated_attention_hadamard.py": "2dc3ea5984f251fb17d93c4868cdc6297801723bc6799ac1a9d4d96bc18e46e2",
+      "src/loop_transformer_quant/rotated_attention_pipeline2.py": "81efddc54396b60dd7c2c412e092e43eddacb0d9641f2d8a4eec9878d72eb4ee",
+      "src/loop_transformer_quant/rotated_attention_query16.py": "161faf38eba65c083fac344853dfade1ec326da7d6257469d10626c2abce5df1",
+      "src/loop_transformer_quant/rotated_attention_query8.py": "143350254d30170998b5300874e1c04a44adc5e3cf23563b33fe61dd1e8c77c7",
+      "src/loop_transformer_quant/rotated_attention_rot_keep.py": "2f55a8ed07a478067ae68b911ee457b4f0628beeae078b121eaae0c8cf5d14e8",
+      "src/loop_transformer_quant/rotated_attention_rot_transpose.py": "dd7823da1e66c88a4f7b74874d25d94261cbc17e85fb02a73938fcfce6fb7765",
+      "src/loop_transformer_quant/rotated_attention_unroll2.py": "90bb8c89664acff921d0369083aa7f53ed628b0168ff5d2e8779654fe727ea97",
+      "src/loop_transformer_quant/rotated_attention_unroll4.py": "4cd2474e5a63dd893d3fc545d12fdf0ea2d2be0e4d232bed1130b5aa7ce95539",
+      "src/loop_transformer_quant/rotated_attention_warps16.py": "c984459e4fc0d9d1d3d0bfe8ec41f1abb7e9ad723da30506ba44a68ca25085f7",
+      "src/loop_transformer_quant/rotated_attention_warps4.py": "44401f179631c6e6120b2e111326e77c14a2edd002ec28d3f8168d102d22695e",
+      "src/loop_transformer_quant/rotated_attention_xqa.py": "e7a8a71abac60a1865a092d25777a499b4fbb6ba8722430c0d4e705f28f4f786",
+      "src/loop_transformer_quant/rotation_policy.py": "7fb06183b276662564c2f64f14bd7ea6a5438b46b3d8a05efc3678f46c4b54cb",
+      "src/loop_transformer_quant/rotation_study.py": "c05e433e8182e7af43ae499d09fc5eb5f8a53512cef4d01be9dd2d59ea26e14e",
+      "src/loop_transformer_quant/rotation_table1.py": "c7bbd247eab272600bcba2219c83160bb54abc4bcff3d1de8ca496fe2faff985",
+      "src/loop_transformer_quant/runtime/__init__.py": "c412e782654be12e77250c6b617b5034c9a2e33a1084b7880f8c0eacb46e05f8",
+      "src/loop_transformer_quant/runtime/cache.py": "2e1f0026738ab93ee3767ca29a53ff38a67d707f3e64f466fc402a80eebc82fa",
+      "src/loop_transformer_quant/runtime/context.py": "e8f723bc270d1293b086d621c66f69a964e570175df733e23c9755e74933ca10",
+      "src/loop_transformer_quant/runtime/module_cache.py": "f189187bbbc1436049ba863550ac0b4134466e09bc1ee3471c3b030b33d01e17",
+      "src/loop_transformer_quant/runtime/specs.py": "b9a3deee69e44c8ccc009e92bdf34c83662b7d553ffebe25eabbead0b897bdfc",
+      "src/loop_transformer_quant/sensitivity.py": "00d0ab71278e19113dd6b5db65d091271fe1be65054562677edf7c082d8f3022",
+      "src/loop_transformer_quant/sm120_fp4.py": "301b7a8141337b68fda40dd9fdeba2d4eac46fbb17c0d61292481910ece83ca4",
+      "scripts/account_recurrent_kv_storage.py": "00aca9766fa7960c010b5170f36781224e45a4a288518bf37c6864b14422fbde",
+      "scripts/analyze_accuracy_grid.py": "0d33123eae6c5beaf625854f6d0b92e4169a835e8fb067139c0b037dcfcf2e46",
+      "scripts/analyze_huginn_anchor_ablation.py": "3d825b07b42a08a8f9dbb8fe7b39bb50c8b51c5c7f02da8183ad11a90d43906a",
+      "scripts/analyze_huginn_full_validation.py": "6b8a5972ff7f878d2368273a41898ccbd5321e61056a59c8ac845fd2f01a7741",
+      "scripts/analyze_huginn_mixed_precision.py": "d8cd6ee6c8f39ec28fffb6ebbdc618443bc88918dd1c5ebe06a8ebac30289519",
+      "scripts/analyze_huginn_residual_kv.py": "015e3bd1abbdc14c98b077d645609ef4fc312654a8c854f421a62e0516d467df",
+      "scripts/analyze_int8_scale_underflow.py": "181ea6a32f88a9edaab237edaa5fcb02d5baa85d518ce996a53ffba214601388",
+      "scripts/analyze_int_scale_matrix.py": "b0d17fc01e36e0cae0bb2fcd12b67e578b0b67983159ee5dbf582cd01a5c7b12",
+      "scripts/analyze_kv_codec_screening.py": "c712a196068fa991a2272435d5c302edd570fd4100209dcd5bf316103557ba35",
+      "scripts/analyze_loop_delta_quantization.py": "dd1201e93c2183a428b1bd0b514a782169d81d03cbec2fed82db67e80f3932e9",
+      "scripts/analyze_loop_kv_redundancy.py": "872e86904e189d18e8440b7f79e80bcbc0061d1b9780aee26c29459710c9d7cd",
+      "scripts/analyze_ouro_selected_wa_kv.py": "f2da633a5e23fa92d9c444b80553149b8644257f479641ef2eec8a9dffd88c3e",
+      "scripts/analyze_paired_screening.py": "777b63be6a0c0b27a07115f94f2da7f50ae8da36818413e6a7b00044895fd79d",
+      "scripts/analyze_predictors_vs_accuracy.py": "0d0c94a94b5c81ab2b11700bbe6a9ceb4d2802ca240d7e1d274954e4ccbee0a0",
+      "scripts/analyze_recurrence_context.py": "2765504af81b1d9025374c1af39d11914977ae1067090292e5a9ecee76a98dc7",
+      "scripts/analyze_run_repeatability.py": "1d529562e36480fa64d32902303c74080f23b5f3358cf04c4e611742df359b46",
+      "scripts/analyze_selective_w4a.py": "ff84fce91a09e11f1d7351a19268086ad9d493e8bdd0657d200fa0128683dff4",
+      "scripts/analyze_sensitivity.py": "5a1d93a6fe3e261f1fe9d54f5535904c378a23b13ef39f482dcea90056c77f0d",
+      "scripts/audit_activation_table10.py": "df38f4deccb5cc437ffe8537a5364277920743c10cdcdede17b1b26709e4e796",
+      "scripts/audit_channel_capacity_reproduction.py": "14558f10129915128726e7535a6f67929e1e71dfc9132865fd94c595ade0b2ad",
+      "scripts/audit_fused_write_model.py": "4bd0788465be0a16fa54709760a5a6ede71d46b1c8128a0f0601bc4bcbb2932d",
+      "scripts/audit_huginn_table5.py": "93913720f049190fa1fcebf93cec2f26d2c802775bb71a5424bd35d227902a6a",
+      "scripts/audit_int2_accuracy_smokes.py": "c0a463d3cd3fea5889c1d758999a3cc059b43c89a57c229513397de6c4521434",
+      "scripts/audit_kv_cost_final.py": "c193155a1d34f2638c05fe65d23b8d68349ceea7586a2acb0f88a8f7cf547abd",
+      "scripts/audit_native_wa_method_screen.py": "05cdf99ca7abbb7333ddce355ca93db4d642b3187fd332007cc5d39bc816d2c0",
+      "scripts/audit_ouro_table5.py": "2966ba1479831d7c941093b5709c00f4b29a7f39a1d8d095025244470cbb35d6",
+      "scripts/audit_portability.py": "465c903e2a1581cc1f84ed7883cecc524295924c31b72dfc685d973460539909",
+      "scripts/audit_quantization_goal.py": "3362e9fc93d2ad3d2aaa89421b8e187c74dd284c6c4a60ad71fe8101505d2086",
+      "scripts/audit_recurrent_speed_goal.py": "7410c511e4ca51c15afedd744b913a2e68fee2b84abab4968e130261ff13bc21",
+      "scripts/audit_scan_g4_model.py": "83acb9495b81dc3ec6d7160c6aceda110826f32ba8f1a50ae1cf5b24d84180cd",
+      "scripts/audit_scan_model.py": "bdd572caf68eef4fd634ce14841512aeb5e2bc64f907cdecbb682d08f0c3a3f3",
+      "scripts/audit_split_solve_full.py": "333caecc946276e9986c9d0a1bb68ab388b4b2b8bbe32b7c1090378cbdd48e33",
+      "scripts/audit_stream_solve_model.py": "11e43ee2ac2f2116684f2818735d980e00613052a06ab53b78993ea92c1cb442",
+      "scripts/benchmark_activation_workspace.py": "1f3c65b83e6b09e211335f0d5710ed9d00533500706a96f36b4bf803cc42c157",
+      "scripts/benchmark_channel_lse_costs.py": "ed77ceb5d1aafc2b1102b874266b6c5bc9e07be8e7ce3e1af14b3a95f0a3fe1f",
+      "scripts/benchmark_channel_lse_policy.py": "362517508242ed94d83e3b11024115be2cd86b71efc7dd4112cbcf0b053a921d",
+      "scripts/benchmark_channel_packed.py": "64be08868ef380c3b905044026db94c289edb1dba849d7aaf74f15d4da6b2d78",
+      "scripts/benchmark_dynamic_lse.py": "0642a3cbb797c802d45bceb60e4b270ea0e793041020320b3489eeff7e14eb71",
+      "scripts/benchmark_dynamic_paged_write.py": "b607ff8de2d0ff083a963cc71d26c7a4d515999e50ebd576353a657d4efbde75",
+      "scripts/benchmark_huginn_native_components.py": "f39093b1d1d339e7fc571db16b2c8c4662b616fd2691a42f35518f16595b4fa8",
+      "scripts/benchmark_kv_cost_probe.py": "6da5ecf0c9767ce784934ee2bee7a0852dd8905a50a921e212520a4d6cfb74c1",
+      "scripts/benchmark_native_activation.py": "ec82e0f1c126b525d49e785f3e085a2ebe769056e77a1a2d13c0aaadc1c755e5",
+      "scripts/benchmark_ouro_native_attention.py": "c5a04ba7798b3631a189187e30bcbfaa08423508f671f08979fa071c445b7364",
+      "scripts/benchmark_ouro_vllm_fixed_generation.py": "92a3d379a73d12350d2c1df1afd98d65146205c412b902bb0051b9a976c4ef9b",
+      "scripts/benchmark_ouro_xqa_seq_trace.py": "d7fa0a4be2e6940e2b0e86071eadbd95004d89b69fe854b05add2fc65cef92b0",
+      "scripts/benchmark_recurrent_kv_chain.py": "085af93b3250eb492823166d7d304ded2ed050dfb42352df80c562c9757f122c",
+      "scripts/benchmark_xqa_candidates.py": "7b130c23368b584b06d3d64ee6dcebd102eff7064ce64edd109bd08563daf4ae",
+      "scripts/benchmark_xqa_costs.py": "ae0b11a31900d21b1e143f5d88fd36887eedcf977154f9a0890870d5fa8fdb6d",
+      "scripts/benchmark_xqa_native_attention.py": "7198b554ed0c10d7d2681c2c297bce03fd085651b552404e4790448d6e8f3cf9",
+      "scripts/build_bf16_anchor_reader.py": "9c072bb017932ae62c2885fcc942c98f8dd7f9e29299151d35ce495e70fc17b1",
+      "scripts/build_completion_manifest.py": "84b8ccc2eafb27025997b981bc9237404cc85ab36f01765101be788eda274369",
+      "scripts/build_fa_tile_attention.py": "5ff13df779225048a1b8f42de46e6a99993569edd51725ab2e3067f0f1bd6ccd",
+      "scripts/build_m1_storage_matrix.py": "a5eeaa9eaeca126e03be047ee4537478f068f5a730616bee8a36558a9f41be86",
+      "scripts/build_provenance_manifest.py": "96eeb4ec1c23af52a5e4ffb415aac532356eb37370c862d5176a4053c100646a",
+      "scripts/build_recurrent_attention.py": "0896c6280513fe2b624c16451e8773007e2acf685adbafb467f89a9127c7f950",
+      "scripts/build_recurrent_read_view.py": "def6ccd89fafa3db068972e426c348e9d29c67d2a4f3506529e3b44c0509d981",
+      "scripts/build_recurrent_xqa.py": "a87ce431cd27ac95476ca6382677fb28093f5942a4a87ab9c7f2d057adbb0a9f",
+      "scripts/build_vector_decode.py": "2eba31bc992c7895b1c17b8588590001617e9707ae2ffc8621f82e572b56369c",
+      "scripts/build_xqa_candidates.py": "4e0e4cab2c5927025f5946a5a79f73f43c3386012ed5b4b95741cd63eb0c7ffd",
+      "scripts/build_xqa_comparison.py": "8fbdad07ddd5ac35df2acf14a5a01a8522eea83877b7cb35ec6a553cd72bccb4",
+      "scripts/build_xqa_cost_variants.py": "88d90a75f931a205dcbb13079846aba305a542e58b0447242194ee39c8b8b5a3",
+      "scripts/build_xqa_hnd_native.py": "a3e0ed25569036b194b0fcfb6f59b8478145396e351cbb0db05dc83ffa2b48c2",
+      "scripts/calibrate_huginn_residual_activation.py": "7189b032c99b5f5c25d2349481cff3f97a4a919ad9510c8e925f97315c7c25b2",
+      "scripts/calibrate_huginn_residual_kv.py": "a5809dae0f75d5c7e45b2453ff1576e3bbf05c1a86f6fcb17fcca03bb69da0c0",
+      "scripts/calibrate_native_recurrent_kv.py": "de867b19016e5cf3d90fac9b1684f13874bebdcb7658e34aff1a059403a0cca3",
+      "scripts/calibrate_ouro_rotatekv.py": "f995a50d5b35adda2560c1f26ea44482cd78145968a07336fbd06868ce50e553",
+      "scripts/calibrate_residual_activation.py": "0c5348f5a0989bf4f178c877b637dae3d9a3a1225e28986760aca39002c5a600",
+      "scripts/calibrate_residual_kv.py": "1b7d6c1c23bbd251dcb7ce3f9f1dd994b581e6c8d8237a0f7c9fa79c950ff9d2",
+      "scripts/capture_int_scale_matrix.py": "f507a18d3394d0b39e94bcbbb45e27cfedb68a9435a3b9fe27079cc365065bc1",
+      "scripts/capture_ouro_kv_reuse.py": "72f411ddcc7e546874aea70e54cee946ddce7e6f85c558087dfe1e2fe7a940da",
+      "scripts/check_activation_workspace_cuda_graph.py": "e0d2c9b871949286a6b0c45b8e5a8b713c6f8f0ad2dbfbfac033084308be2310",
+      "scripts/check_bf16_anchor_reader.py": "107caee9b7a3a58d1554016feca042c545240ce5c4f2685f5edce0d1d783c6d0",
+      "scripts/check_channel_lse_certified.py": "141e493ef9ac943fd65e4b0352c48844ab1e250ee577064d98b4790364ca93c5",
+      "scripts/check_channel_lse_native.py": "7937d8c0d3b94830984ee8fa19c9af7829689ce4ff1ca20f448276d587fe993a",
+      "scripts/check_channel_lse_split_solve.py": "ed222515a14b35c617f80259d91d53b98563a2928009f6126d7a7b93844b1734",
+      "scripts/check_channel_lse_split_state.py": "8634d6c172652760f57070b2ae9a782b3c1fb865832744591f3b89ee2272790b",
+      "scripts/check_channel_mixedstats.py": "eb80353d62f2569b012589290510a165d4a9d04190f2e15c319534c366401b32",
+      "scripts/check_channel_packed.py": "7d5983fd3850838f6da437196f29459f91248cf223900c063f5032ed37195094",
+      "scripts/check_channel_packed_heads.py": "4cbd129906f2ce7a5f4e91eafb50724bf49eda9df4eac126531fbdb80b342c1a",
+      "scripts/check_channel_packed_split_solve.py": "75900d0a014901182629864c5575ee0d01bfe4e1b92b3dc387535b26560973fa",
+      "scripts/check_channel_paged_fused.py": "72bbadb9214c5815050f1c3436659a6619632c42d6bcf10d4cadca8665a7719d",
+      "scripts/check_channel_request_reset.py": "eb785b1301ce8380eb7052b4433c63891f7d4bd14a369125af1930ce2a5ae4eb",
+      "scripts/check_channel_scan_numeric.py": "11c090307331ed7078c0986e3c5436bf1b5387f078afe78320bb9e645d5d429c",
+      "scripts/check_comparison_attention.py": "fb8b20d36f57efd3eae00db870471e285eafcc68bfaf91ab9a7a54d119aa0571",
+      "scripts/check_dynamic_lse_recurrence.py": "7d6ed664c3004ff6445ccf9f6d3d7c4501fe346a6e5b13a745e159c0855e669f",
+      "scripts/check_dynamic_native_scalar.py": "fcefa1178f06cd7281c9fb67c858bf70b6857afce5699b3617f11b89a97bdda7",
+      "scripts/check_dynamic_paged.py": "b42960a5f7795f13646e98acc937b810b9f1e12cc1b9ba0d8a0d00a3feae12bc",
+      "scripts/check_dynamic_paged_fused.py": "cea6ff572d72bbfb4f03a5b0325eeb6bf17758c58f052487d4b40c3eec1ba456",
+      "scripts/check_dynamic_paged_heads.py": "c17a61e56ba883cb92b9aa54a5d8b557abd337af32cf9711b90979f7dc0b042a",
+      "scripts/check_environment.py": "00343fbd026557b2c317c26883c37d0218495512cb38bb99506ecb1c744d5490",
+      "scripts/check_fa_tile_attention.py": "1b23aa2a0d25c3176c175c8e841e616e112ff40e70a4c1e4178ca881e629d9e2",
+      "scripts/check_fp8_read_view.py": "5d2cf0e70bde90177c5a6be7234c6bb94fe303957e595ff2f7ea1858cab49708",
+      "scripts/check_huginn_graph_parity.py": "e1b5002ed78f7ba49a23e62db86f29841365265330c5d5cf3bcfcc87391d8695",
+      "scripts/check_huginn_sm120_fp4_parity.py": "6916213489c904c13994ae663b7f219c3890d45c9247353600e66929d3c3462e",
+      "scripts/check_int2_experiments_gpu.py": "400ceb0731884a208d3b5225fb4994e682a92a0163ec854a308c55c9fd62b091",
+      "scripts/check_int2_streaming_gpu.py": "dbaeb741a1215f06de00b78bc0f71c6e6c01b8a100ad9c180b4ca0f6848c2b5c",
+      "scripts/check_math_modelopt.py": "7924917242ed769f80ce00d18043ef7b719706c480259182c076f98ba270b62b",
+      "scripts/check_packed_affine_kv.py": "acc10f3ddaa6b38957c42f0741301668700ad2882711117fa990d574feb53b20",
+      "scripts/check_packed_fp8_conversion.py": "b05e9ef6f321715f93341d079c4a0c13cd1c516763d91be4b55957ca02016314",
+      "scripts/check_paged_comparison_attention.py": "c35a2f8983149403f6f13986ad38e4e3df331908cbee5ffafb70eeca75268359",
+      "scripts/check_recurrent_attention.py": "18d513c8b2545f7c0f393b5edbdff0451fef22bebb6f2a3b6ca7c90083090179",
+      "scripts/check_recurrent_dequant.py": "3717d5bb5d012a73e3c51416330d82310e6baa81c7777a32277920a78b890e2b",
+      "scripts/check_static_paged_fused.py": "69d122d8d6bd49728990ea71a5437a1ddca2ce8a2d6e01e3e15cf067f1372e55",
+      "scripts/check_xqa_bf16_matched.py": "34a5a6469befd2f372caad2b278d1ff1d9b0b06b842e63003c36c81927acbe40",
+      "scripts/check_xqa_fp4_lookup.py": "5aab5b1a773ef3495e65e69cbeb36657d884b1cd0567c831285c91f747edffe4",
+      "scripts/check_xqa_fused_prepare.py": "2fb54c3928dc77e369887517396bad325ea44ec019c2ec29bca9c0f85925831b",
+      "scripts/check_xqa_half_product.py": "b55b1632d22a562271815e55250f8f6ca1493ae8251f1fcfbe56e14d1eb8e61b",
+      "scripts/check_xqa_mixed_attention.py": "b0708b1cf369377d5e65efd637b2126bef1e33d5be7940e0b2d7fd6f990a1de4",
+      "scripts/check_xqa_paged_attention.py": "9d0fad47bb905da01d8c0eb2d76084dedac321bf9771fc349de6ec42b346d870",
+      "scripts/check_xqa_paged_generic.py": "f3efb933095fd47da14c64e86d463dbfeb4e7731c5e0dd50cec2a6bc8f2ae15c",
+      "scripts/check_xqa_prefill_decode.py": "bcb51bd5947ccc1ca2c490ca8a17dcfdbf1c68312a6a441b724ddfd1998184db",
+      "scripts/collect_activation_momentum_payload_heatmaps.py": "e773e670b76f5b5a0af600edc340535f028121d122daf2bb3d5c50c13aaca2d0",
+      "scripts/collect_gradient_tensor_norms.py": "a1d6d20ae81e700badf807e445a770a24a2f2dd2cb50c7659dc07efe910fa031",
+      "scripts/collect_huginn_anchor_mechanism_cell.py": "b9c58fd7fc3fd6caa8ee5875f4a8619ec736646afbafcc4156285e15d0caa055",
+      "scripts/collect_huginn_layer_channel_heatmaps.py": "87977e4d6a34e3422fa42af1ac943cbd17814eb6911f8f19297318b72ddeeb6a",
+      "scripts/collect_layer_channel_heatmaps.py": "41a293cede5636d07ab52baa61d7c495b63cfde5915ad9daa8a15db62aea99d7",
+      "scripts/collect_momentum_payload_heatmap.py": "0b94a57d71fb7b8318c21f9f2e6fc2e604317debd441a1a73fc5d0dd76c28c9e",
+      "scripts/collect_ouro_gradients.py": "d7b7c0671906c75c353ff34b257b60ccd5405479647f85bdf957204834fbf493",
+      "scripts/continue_after_q08_uniform.py": "10c0b22748aa4066bd69c49f9dc4e9c4ae50068fef2e1bd7b953bce4568412eb",
+      "scripts/diagnose_channel_prefix_cost.py": "f29704a8cc5d3fd3ce81892fcb06395349d15ed64a32d41f2936be8a18121e6c",
+      "scripts/diagnose_int8_scale_underflow.py": "a5c134d657da16dde0cab5eeb09fcbd3e3c2b837e8ea8447defe0354f6546b40",
+      "scripts/evaluate_gsm8k_cache_codecs.py": "2bf64cde6944cc6cf3ac3e88ef8d22336a27fc4da6c2d7b4505c0fd930a275cf",
+      "scripts/evaluate_gsm8k_fake_quant.py": "22d0c237f74310ed2a8de0973b8bbbe59bee5c57ade8e5fa35f51b7e941d0b8b",
+      "scripts/evaluate_ouro_attention_precision.py": "8fc0f383cae171c45fb3029b83ff42bdf0e0e37b761c3e6e58abb4dd30ff9562",
+      "scripts/evaluate_ouro_code.py": "649415991ebaa7894c40e0900c76408ee2fbe27a6a483f5438cbe2d1d99712ff",
+      "scripts/evaluate_ouro_recurrent_vllm.py": "d0f14e137f425dc1e0f6d2341dbefab3f4df2d04a313760ae99e2b2ff80bee2b",
+      "scripts/evaluate_ouro_vllm_attention_precision.py": "b945f987d6dcc98a25f8fd2f1eb307774e3d4b6cb1170c5e2b2d981fa54b9ba9",
+      "scripts/experiment_manifest_common.py": "49b5ca26f2087394144ddaa6d0a36b07f16e503b7a838a382fd7b8f2257b74e7",
+      "scripts/experiment_queue.py": "21df177893db181b87b13dfc04d9e5538e81cb4770cec8225df7485d91c0badf",
+      "scripts/export_codex_sessions.py": "212d5d0428d31aca43345cf6017ebce23dbfa98134ac85ab0b1a8ed16f2fb0b0",
+      "scripts/export_native_nvfp4_checkpoint.py": "b7373493e7a13586827c370f8521e544dcf238a16792c89ac60174c8a7eb77e9",
+      "scripts/finalize_activation_table10_report.py": "4f9bb32ce670eb61f270e9dfcfffb7786625c5e73c42b111e0c66d792a1eff41",
+      "scripts/finalize_huginn_table5_report.py": "dc1994ae56f9100a92c4bc5417c10bab9b7d148d81200732740acb2522ad17aa",
+      "scripts/finalize_loop1_dynamic_table15.py": "753e956cb404d2401cfca8928eaf5d8cd568a9d17c1b93971a10a5ba2e83123c",
+      "scripts/finalize_loop1_remaining_resumed.py": "94cdb86481d84ddcda7abcc7831f9b3a93da857cc41cf3e592453c8369fa224c",
+      "scripts/finalize_q08_uniform_bits.py": "3c653a6ac63e6495e1c293fdfaee85c302cd657d770457e0c63735cc5910c752",
+      "scripts/fit_int2_accuracy_candidates.py": "bb9d87115ac4213cff11d12e96d83e1e5c4eade0ac67821fb5b3df326c9361d7",
+      "scripts/generate_activation_table10_tex.py": "f8f855ba9870dafab119fe9ee3da4d085814f36270c1286296a2ea4d339a00ca",
+      "scripts/generate_huginn_table5_tex.py": "7253a086a2cc10630211307fec2142c7e12ec06e010e0c64662acdc0a845f11e",
+      "scripts/generate_kv_codec_report_assets.py": "9f0fcfd69147f1bc7feb3a54b9cbaa0c4465bd73eeb92b4d8f8fc99008dbb975",
+      "scripts/generate_notion_goal_results.py": "c24ce41bf220cef7f3bbcfe3b0b656a9fce58c28a444f71506f741e683ecbb3e",
+      "scripts/generate_nvfp4_residual_report_tables.py": "830a8a6c038f739f3dc44c056601db515fe82f3df94c118dbf625dfc190d8dc0",
+      "scripts/generate_precision_flow.py": "24b713b71784ec3c84f52940b535d1027bd5ca40fb73f81baaedbffdcf7a1bda",
+      "scripts/gpu_availability.py": "09c716b2fad2e969b3eda9acedf91817d4dfbf0d67280563bfd2588386a486d3",
+      "scripts/huginn_table5_matrix.py": "22eb8f679a832ecbbe6f2d43ddb9653c2103ec3417cbd246b56e33abd0e59325",
+      "scripts/measure_sinq_nvfp4_weights.py": "cc2b80b416b5c47b415bb4c60ece87d394ab61b751fb726c206abee9c7c11d58",
+      "scripts/merge_accuracy_results.py": "a8d0d79613893024382ed874b497fa08154bd1ffca88e3bf8ff669216ef378a0",
+      "scripts/merge_full_validation.py": "ea7182148e8b98defc87b664176570a75ab08f9bec5c05799e70c25b635cbf53",
+      "scripts/merge_gradient_tensor_norms.py": "850ecd33e3fe3539fc83b6adb8232d58a8e3f6a540ec1c07c4cc17ab6b38205d",
+      "scripts/merge_huginn_anchor_mechanisms.py": "8d02dce1f8914cfcfe18118378dd10c370d52b619e598bc181fb65d6e7b431fe",
+      "scripts/merge_kv_codec_full_validation.py": "579713ce35513f10537b2c3abba3f7b33aedc05d73aa76c01d103c85adae48cf",
+      "scripts/merge_layer_channel_heatmaps.py": "5a56b2407d5ada4ab61415e00f25c4780926d1ce2bffd03707576eda52d6cfd9",
+      "scripts/merge_sensitivity_shards.py": "1b1afa38007afe6bf4d4e54e211cbb956b3b6af79b1fe0a24300a33c99851e06",
+      "scripts/plot_delta_ce_summary_grids.py": "7029fe408fc3e2e8dd29407d22c2c6b8fa574b8a374827791704c71271282672",
+      "scripts/plot_gradient_tensor_norm_heatmap.py": "f888343b48a10cf142532c391768526cae058bce7a70197453dc1387715b0fc3",
+      "scripts/plot_gradient_tensor_norms_v2.py": "7a057b1c2b3a7251eb8ea565a64a32a82cad4bde00abd2b4c6ac53bb0d1ae00c",
+      "scripts/plot_layer_channel_delta_ce.py": "c7167a3a26b92e7804c4594494e0e102df57fb3293efd15a703d97e37e9c3c9b",
+      "scripts/plot_layer_channel_heatmap_grids.py": "f0ccfeb3f0dbaa6e3e7710b2e8e36a5dbded48565aeae1dd78565c1fafdb4d0c",
+      "scripts/plot_layer_channel_heatmaps.py": "e436e4e3b24945bac35785d0c2882ae4ef253d9ccb88281333230ba0275ef601",
+      "scripts/plot_quantized_payload_heatmaps.py": "8b652720d49a64b32336f09b3b7d72d21b754d6c84d238a821c612ea9328d37b",
+      "scripts/plot_quantized_site_gradient_heatmaps.py": "89859986a1625c6143a562eb6441c99760c7a310f287f95cf82aed0551a72ab3",
+      "scripts/prepare_huggingface_release.py": "a0ad51ea23de34f4c5023d61eb6afe23c9ac65a4c1d151c7ad574ff8c3614326",
+      "scripts/prepare_huginn_pre_sinq_checkpoint.py": "2ea485abd7d3f9458e243a768d8e4fd54ccd0bf23b2b98489608703533a1bd2d",
+      "scripts/prepare_ouro_pre_sinq_checkpoint.py": "923838b1d95298e9ccaf81d11e4dd39c6f4a7834596e354075f7d5ee60253be4",
+      "scripts/probe_split_solve_vllm.py": "e8d8b3ac144dd8c9b8574f3c0db35f706a1f97a1153bc3b7bacda4a14fe04f8b",
+      "scripts/reproduce.py": "7479a7e865d62d16d245745efc964e532eca419ded3aafda7d2a4951cec88697",
+      "scripts/rotation_study/after_hnd_memcheck.py": "bbc9751a541c33a3f96bc2d3b5d14657e924b3207ba96a21613ffa69b9797c11",
+      "scripts/rotation_study/after_microbenchmark.py": "e0ca715d17ffb68bb61671a19250d546b2165be15ca903db70a2448b8b1685b4",
+      "scripts/rotation_study/after_resource_audit.py": "06f2c32b4f0cdc6ac7dc91e1f7b529c731cb3695af97459b9171326898b5d8a9",
+      "scripts/rotation_study/assemble.py": "59f0f0fe62e17f54b63deef52bc8c6db3173d05d781632a055f58f5f2271e642",
+      "scripts/rotation_study/audit_adaptive_safe_resources.py": "e9c9a60429e83b33391aa9451ca452ed2a0b0f10b998c498ea580903c2f47256",
+      "scripts/rotation_study/audit_combined2_resources.py": "949ea01ea47dada23e561352ad10fdbca1bcad40febe89043b39a73e6d9f9e10",
+      "scripts/rotation_study/audit_combined4_resources.py": "5974bf8fe837f653f01e72aaeb19e5a5b3d5da82e3e131db136e5e1fc699b0bd",
+      "scripts/rotation_study/audit_extended_resources.py": "23522c6deab12c048be5398e26d13e0286dd58d337165e4e82bf129d5883e1c5",
+      "scripts/rotation_study/audit_extended_resources_v2.py": "c1d7d7d2dc185726c60a5fddfa54e98da9ac59fbd146a7956e03a323c79a14c4",
+      "scripts/rotation_study/audit_kernel_resources.py": "0e75d8bc8c926167f2bbb80b929aa2a0ed1987652757ea5735062411b11f10e5",
+      "scripts/rotation_study/benchmark_attention_conservative.py": "0738cf41aad38302ccef140f84a5ff4c7369811f77ce10f8a65dbf538d7f6a54",
+      "scripts/rotation_study/benchmark_attention_fast.py": "7ab5657d2d65600f4123764fc2fa68710a5bcee7ea9d48315a159f582166181a",
+      "scripts/rotation_study/benchmark_attention_group_load.py": "7731b1c4ba41a4c1233d9affde6ff390adc52e8694cba34d58191e6d90c2ed67",
+      "scripts/rotation_study/benchmark_attention_hadamard.py": "c9dfcf998b9966f24be6bb28beca5860b1c5f3e3eef2d39c13c7aaa17e08b76a",
+      "scripts/rotation_study/benchmark_attention_pipeline2.py": "72ade1025ea6b7a7cd75c16b9dc511288c0457af1f17650ed9a28483bd7c568f",
+      "scripts/rotation_study/benchmark_attention_query16.py": "50ec0df96debf7efcb79be22794d9224a41ef8a6661ed1934fcf700dbcc9e4dd",
+      "scripts/rotation_study/benchmark_attention_query8.py": "b694221a3443dc5678eef6fee96001451d1b0b6b8a8dec4c13d6353c23459e9a",
+      "scripts/rotation_study/benchmark_attention_unroll2.py": "a33454010fd4476c29668f9b8244303dffc6e1f809c4fd1d84b8648d0b944058",
+      "scripts/rotation_study/benchmark_attention_unroll4.py": "75aeba671bd83fbb51b9ccb53afffe3cd5a33522653ddc488974de44ebc1d5b4",
+      "scripts/rotation_study/benchmark_attention_warps4.py": "b9b5ae3e8e6e99c6749624da82a13cec9242289438f6fd9cb2f60f71ba050f95",
+      "scripts/rotation_study/calibrate.py": "229bd2299d8d2fc581c9070788e63b75b991e15fb1d713952fecde6a3f888dad",
+      "scripts/rotation_study/check_attention_combined.py": "bfdaa9f9f05064053372ae3cecd3b111817676ddea88ae4d7a50a46adb74b9fc",
+      "scripts/rotation_study/check_attention_combined2.py": "102fc9d461c92391debc06443f045c863f935ef3526b0edb8d99c413c25ae17a",
+      "scripts/rotation_study/check_attention_conservative.py": "3c618c8acf3761c91fc03162f76209f92875b909bad12f9d240892150eecef33",
+      "scripts/rotation_study/check_attention_fast.py": "960716ba5ccc9fea0daf7c2b5b7807862a30a1ca6190096969b4ec98cba88a88",
+      "scripts/rotation_study/check_attention_graph_growth.py": "e13e721cf4c77e319832f079d37d84c0499cb8096f1a8885a2dd4224b2b5cf3b",
+      "scripts/rotation_study/check_attention_group_all.py": "903fb7ce4ef652b9a83b3f7dc670b7eccd81a4c4c27db4c036e7f35f7c561df8",
+      "scripts/rotation_study/check_attention_group_codes_v2.py": "1c7adde1dfa15bda45403f9e3280c97bc41c2917c41cb3a4b09e1e8b57bab1f3",
+      "scripts/rotation_study/check_attention_group_load.py": "e0df7bb939e31299536af8711816defb19e4d6c0ed9922d9ad3d5338c0cbb6dd",
+      "scripts/rotation_study/check_attention_query16.py": "7234778ff9fcfa4682cb8c473f2aeda2596cf05564b5c5108d9a0f8028a5a058",
+      "scripts/rotation_study/check_attention_query8.py": "67cc348b035731dce563728a7c502d980ebff9c218ec87b06c8a082921f1415a",
+      "scripts/rotation_study/check_attention_xqa.py": "f1ea76d15c447ae926daef4532c86de9f74d14c9e2c7ec6ab731439869ef20e1",
+      "scripts/rotation_study/check_clipping.py": "456503eb2ceb13231f896e6bca8bae707d0fb8c0c34f3bfbaa88813089044766",
+      "scripts/rotation_study/check_packed.py": "75ce77cc5689f0d580418631b1b7d651db88d67cc2214af80b56f1d6cf839bd4",
+      "scripts/rotation_study/check_packed_extended.py": "a61f4df471eb030553b5c3845773743ca4ca90e956550d4267bac289aa8f70a9",
+      "scripts/rotation_study/check_reference.py": "96a1fa527711697f34ad9f9b9721da53e842cde85f9799c64203a022721fd328",
+      "scripts/rotation_study/collect.py": "175fb6af33c49bd09ef5df3a6329aebc97a2666c5b605442125cc930f1eed766",
+      "scripts/rotation_study/diagnose_hnd_alignment.py": "f49d2ff32017adeb6ed60bf718f7397e74e83bea0d15b5f064fb8d58112f3339",
+      "scripts/rotation_study/diagnose_hnd_alignment_safe.py": "f7c3e64765b0cc6b86b4e374e46485c92d20fef33bd6a28984bb1a7e1172d7db",
+      "scripts/rotation_study/evaluate.py": "32621676094335cf3d96eb81801d72288d9d6390a7329f9acf9c06a3aabf3a01",
+      "scripts/rotation_study/evaluate_combined.py": "1d755d82819038b94c3376ae6f9c06b6cc8e33119f8aff3722ea4de9bc8fd973",
+      "scripts/rotation_study/evaluate_combined2.py": "683300fe00fd9a2bef03c7bce257ef7a290e032e901b5bc90fcc5fc92f1f45ab",
+      "scripts/rotation_study/evaluate_conservative.py": "fbcbb6cc82cbfe1e16515b3f5f068ab356ad214ee37bfbab7adca15da39eccf7",
+      "scripts/rotation_study/evaluate_fast.py": "4a6ccfac980e651f6813deffcd8ea14d435186cf29c4386e3ba4041b169f434d",
+      "scripts/rotation_study/evaluate_group_all.py": "ed9a980dfd7b2140107ba2d0748940df49cf636af0d85abb804b82838c946216",
+      "scripts/rotation_study/evaluate_group_codes_v2.py": "e5e24323924dd11083681d5a9f74ceec40677463ba9704ab8d479554dbf530fb",
+      "scripts/rotation_study/evaluate_group_load.py": "6b5f72d640017a8787d4f4c00d4a155158b662e073cc68f3585de7a8a6d35d25",
+      "scripts/rotation_study/evaluate_query16.py": "cb3d9282444f2c2a388ef16a7875a88c69c88f8a4bd4314bd91212e2124970ec",
+      "scripts/rotation_study/evaluate_query8.py": "b14553e09b497cbc9b64b7f12ee1d8770ae1db01d1efadc1f1afd0ba39810e30",
+      "scripts/rotation_study/evaluate_xqa.py": "447268526c0b2a146a59a0ebfb81fefa5e7120050322f95e2ca212ac1bca3074",
+      "scripts/rotation_study/finalize_recovered_study.py": "0659a2c8cdf10e7f297356a6f3cae744f63c6f6494fe9167314946c2c77470e5",
+      "scripts/rotation_study/finalize_when_done.py": "0f97ba9270afb4782d699815c0afca65e462a363fd0f626f0733bc8a81287e90",
+      "scripts/rotation_study/pipeline.py": "3e79d4d7fbea491ea361def4ce86c2077871151bc8ac8b4d2c44664ac3eaad14",
+      "scripts/rotation_study/profile_attention_costs.py": "91ec37ff0c5e316c68aa9f449f696da41e9e1facebad4a6e03f61954b7e4d6fa",
+      "scripts/rotation_study/resume_full_cells.py": "fe0a80d87a046343119fe3fc0771ba8597235d1879ce8ff0f60102fd1f87bd42",
+      "scripts/rotation_study/run_acceleration_comparison.py": "2edb6c1f4fb5d6d7900b0a26ab3e4a1890f8fb030527648952299b8cc62d5771",
+      "scripts/rotation_study/run_combined2_comparison.py": "f344b81b7e96f7db40a49a2f8f52da685f699e9005f43df78da7a5a2a1d1be94",
+      "scripts/rotation_study/run_combined2_followup.py": "d403d837d6d62d4d6c50fc0c3e06ecc28c273bc0842966c3a312dbdf38915cbb",
+      "scripts/rotation_study/run_combined_comparison.py": "a4a276513fa311a44488d0fdade035916d5a8a238ab41a619d7152b8a6585c42",
+      "scripts/rotation_study/run_combined_validation.py": "c8d22d1f36189ba7d8b77b2284b5bc1d1b71815d4496ac0e4b92eb1681db5bd1",
+      "scripts/rotation_study/run_conservative_comparison.py": "61405a1b15adc4d3c0e0c7db08897b98300d7cacd7183fbbb34a8da63db25412",
+      "scripts/rotation_study/run_expanded_acceleration.py": "e89890efc49cbc49f544c80e6e82fcc1acca7032bdad038a5e466e9ce8fa0f1e",
+      "scripts/rotation_study/run_group_all_comparison.py": "6dd53bc538e3b377535ef83bbc344e64db37a3d7ebcb4cb6e276f9f1a6251186",
+      "scripts/rotation_study/run_group_all_followup.py": "e42d4e2230db6a02f589d0d8a5d9dfdb09a24a2c5d49cc0a063f88407730a3bb",
+      "scripts/rotation_study/run_group_codes_v2_comparison.py": "129de34774e3191999e2c75fea8990531b4a8ae9ed8392d935710ac18214e177",
+      "scripts/rotation_study/run_group_codes_v2_followup.py": "457030747991052eb77fb3152cbdca78584abdcbbd1e5cfff98d22450c0090ff",
+      "scripts/rotation_study/run_group_load_comparison.py": "61e770792b1eb50f5bc76cc2f8368611839fe51353e84b89366b07935fa215d0",
+      "scripts/rotation_study/run_group_load_followup.py": "c23dfc84017bbe3d7f84e418bb51e505335d732907327a1ccf2a44f9e1387af7",
+      "scripts/rotation_study/run_loader_screen_queue.py": "825bbda7fbc6549ccf470ca958349110c0487d43a752871cd70a4a563dab710e",
+      "scripts/rotation_study/run_query16_comparison.py": "ffcaac8aad3c9db92567e4bef208a484dfb4063781f8732d3ae16f7c581ee9f9",
+      "scripts/rotation_study/run_query16_followup.py": "e8a7efda0d5a50e7e87d939e90cb755485ef4975a2295a4710157ace4284ffab",
+      "scripts/rotation_study/run_query8_comparison.py": "594aecbcf665813567d0f5c61c675504da6cd77f241f597f856f5f60e500516d",
+      "scripts/rotation_study/run_xqa_comparison.py": "ecdb552092650d2cba17a36e8fbe736d67c5a08737f15183ee3faf5f55ec942f",
+      "scripts/rotation_study/screen_adaptive.py": "c47cb8699940ab6ed5c822ec203d73545dfd290d1bbf01d9657ff8cf9ecd0a79",
+      "scripts/rotation_study/screen_adaptive_safe.py": "d40993ff4e40a9a617f54c35165ef52c6d3c58146e8117f112992b3433aa469a",
+      "scripts/rotation_study/screen_combined.py": "bbca232a238a01f9bf54252723980dbbe0c132d684e551334262be4f956f9fae",
+      "scripts/rotation_study/screen_combined2.py": "b35609da38f021ce247f7cab965e8dfa0ea7a2be9abaaa1c8175cd9e1f5bb840",
+      "scripts/rotation_study/screen_combined4.py": "7912e85e5dd2cd10ba2b87e4099ec5d50e1b7229f0ad024f0fe3f5afe6005b59",
+      "scripts/rotation_study/screen_group_prefill.py": "8b50b8a773174e37a1467db8cc7cb9d25e9263f8fc0239ccdc4097c8f8cb360e",
+      "scripts/rotation_study/screen_hnd_cache.py": "d208cfcfd036325d7d1a8b406480fc1f432f3833172b26426bbf124d0d34b6f6",
+      "scripts/rotation_study/screen_hnd_cache_safe.py": "f7251a4dbffeda1f465bccbd48f9ee201e10d461edb0998c748abdc596752eb3",
+      "scripts/rotation_study/screen_loader_candidates.py": "d201f41d4bc5f5c5f33f2fc62c601087f2eef629f1e1e3e4a87a4732891ed3b8",
+      "scripts/rotation_study/screen_loader_candidates_v2.py": "c2aee9b56fc05510f38880f2072a12b65efc938d67d5fe391cdd6a9a8277edf6",
+      "scripts/rotation_study/screen_rotation_access.py": "5cd4cbd0df0aaf9d7c8e154bd4a206ed5f1f9b47155659dafaa87641d4d71a54",
+      "scripts/rotation_study/summarize.py": "54be9e99168c4cf430cb27ba10de54fb6d6e35268b4466c9b45b0ee63fd246b3",
+      "scripts/rotation_study/summarize_expanded_acceleration.py": "1d15778f3b1e093ebfd8b4deb8fccbbe99035752ca691f2b0d0be445a28ac933",
+      "scripts/rotation_study/validate_acceleration_finalist.py": "1432e674519f118fb26ead5ba5d1396b80b3cb42e4f55b95be6d90a4c6a389a6",
+      "scripts/rotation_study/wait_validation.py": "26265d2d904c2734be7fdc2a518ede8b51493d22d7785b332d32c0d77f383aa5",
+      "scripts/rotation_table1/assemble.py": "85022a1d7ebd6758206162ef7eb0420b580e2c33c826db908892e8198231b680",
+      "scripts/rotation_table1/assemble_identity.py": "07c311f0589b47010f958add0defbc9b789af481d92e99098259cedd744bf528",
+      "scripts/rotation_table1/audit_smokes.py": "537a099a365dc36f616059887e280d369c363d4088aaec88780e77755e0fa412",
+      "scripts/rotation_table1/calibrate.py": "bbf40471f159f7af193362271aa4d2a7b6ce34e8a32959abc8fc7983edfb0b1d",
+      "scripts/rotation_table1/check_math.py": "ff0b97813f57eaec01035859d178c5b8aeb9828ae34008acd32731fc363e7fec",
+      "scripts/rotation_table1/check_packed.py": "57aa4980c083806bed498a2025f3e4f868cb03a6e35867f4cc722d4dd7bde961",
+      "scripts/rotation_table1/evaluate.py": "37b81a409d5bf8b9f5449839022d98059fb90c851c19c7a19c32f4599a7ae902",
+      "scripts/rotation_table1/plan_table3.py": "d6f579cab190f6442f5d6fd1b74f61d5244a1a78698eeaa76e120e01b5267212",
+      "scripts/rotation_table1/render_table1.py": "eb9d6e548fea8f753f11475e80146b4cc7b535a76a24c2d67bca9fdca42660b3",
+      "scripts/rotation_table1/render_table3.py": "5bc33b107c92760cfd6c3dae3b0dd7e8e72cb1e68533b17a44208c82081b40ab",
+      "scripts/rotation_table1/run_full.py": "b4bdcf7fb41d4e79525bd146a0fc589792f721769cb6441e7a17a799d4c475e2",
+      "scripts/rotation_table1/run_smokes.py": "d216bf8354b3c9d18bc90ba103a9095a39b80c8c20acab27656ce1d7828fbf5b",
+      "scripts/rotation_table1/run_table3.py": "1ee705fa75413c69fd0ee11a44029cf2c1bb511c36cc9858f2cd65bb35d1f258",
+      "scripts/rotation_table1/select_method.py": "c1d2f45d1f7bff4639e9c12211889076289483176a84435e3a9dadc7a46f572b",
+      "scripts/rotation_table1/summarize_calibration.py": "4b5bb9bd7d07cba440da7ce89bf365de32ee8b6d2843aa23dc4969bbb85d479c",
+      "scripts/rotation_table1/summarize_full.py": "bf94aac5c5ccc07cbfd82e38f01834e6989f386301f45fc8f378e177f688f81a",
+      "scripts/rotation_table1/summarize_table3.py": "1323e4a7cb40fc13862d1b455b241f241608cb151b72dfa629363f540cd843f3",
+      "scripts/rotation_table1/table3_jobs.py": "c368ef161c9bef456a39d215673f7b0f98ae75cddc8363cbf1b3ce1dcf533faa",
+      "scripts/run_activation_predictor_matrix.py": "5e9756e04612221cea145990d118443c3e747d516bd1ee3ce21d4e0403ca763b",
+      "scripts/run_activation_table10.py": "be359602b5e147ddd0d28a8f28726dbd9b857622bcc037bc7843e91caee992bb",
+      "scripts/run_coarse_sensitivity.py": "f327fbb2e2b80cb98d6b00b49e96484dbce1f479bc67d4e33b1b6cc9ff2d16f8",
+      "scripts/run_dynamic_group_scale_multi.py": "69d91faf446b9103d4c00f5175ec1a0f786b5e719bcaf3c9f94872dc87bf01ed",
+      "scripts/run_dynamic_group_scale_sweep.py": "4e8588a8f1ca7f02b566d8ede5373a18cf56a014d84487d458bcf4b7c2d64b68",
+      "scripts/run_fixed4_suite.py": "46eacb9ddfa89466cff33a1610d2051e53dd22a2489c933008757c1ba3eb8605",
+      "scripts/run_gsm8k.py": "594e570e58ea79aaaf27e6fa20088e4a63856044cf292b3d8aec57e658ea31ae",
+      "scripts/run_huginn_anchor_ablation.py": "d249e21f49f7df8421006fb847b51ad4848725ec0b7c16f4e2601bfb3599793e",
+      "scripts/run_huginn_anchor_ablation_queue.py": "544e241d06cb2aa34ab7a5946c453ef184f64fda5567689989b2a2537dfb3b65",
+      "scripts/run_huginn_anchor_calibrations.py": "fd4f8909b3fdc08745a5f26021096f52999fbed7cf510f06a6c4135acb0907ce",
+      "scripts/run_huginn_anchor_mechanism_queue.py": "b201ca02b2f18849bba4c05103fe5323d428cae4bbd919c8db268c1bcee4847e",
+      "scripts/run_huginn_full_finalists.py": "c6d79cc94e4d2314c7499f785cc196ed2f31cbfdf12d19ed18fcad8704bdb941",
+      "scripts/run_huginn_heatmap_queue.py": "8958642e837aecd51f0eb1f5d05b873974983ca24fd02779e7a03cf94a914a34",
+      "scripts/run_huginn_mixed_precision_matrix.py": "f64249ff45a0fc0a1813bbc0960bd02b1f98c87e7571dfd697d79822f1b91286",
+      "scripts/run_huginn_presinq_native_screen.py": "592d3a35f85274ff3ef15386dc348f300185409c27c5673c81d54b525a9998d2",
+      "scripts/run_huginn_prespecified_full_queue.py": "266936101d144add1e56290205e01c4cd44e82441e303334402a388b1ef11a07",
+      "scripts/run_huginn_quant_suite.py": "77b3d88fa71b5ccb3710f844da84b2c0e1a87f572dff09ca90c2e311cfeff0a5",
+      "scripts/run_huginn_residual_kv_matrix.py": "3fc23607e30ca800eec5b4040d715a2746f3f3011e627d5cab9329315185e387",
+      "scripts/run_huginn_selective_w4_screen.py": "74f7c15da787e1b7a4c001c5ecac9ffdb6f5f1c9fb62da6d4c1cef1cf5197c7f",
+      "scripts/run_huginn_selective_w4a_matrix.py": "f1ee23f016b8704caaba0d09502e1a6cad237f5b99c912ca2c44242d87d7fd86",
+      "scripts/run_huginn_table5_queue.py": "cbd005233a00c9dc56b8b866845236a47867cc505e2009f47d8d89e914996a7a",
+      "scripts/run_huginn_wa_matrix.py": "1fcc70489a2cdc0a6db05c515b47699e0d4fe49939ab5f0fd8999df41f264502",
+      "scripts/run_int2_accuracy_full.py": "ace6af465d793dd1f33a922fc0de8ba8ac832c519f3f61dc91992b079671be93",
+      "scripts/run_int2_accuracy_smokes.py": "47c987f932ead87cd8715fed7df8c09f6bc18c9385c5d31d5f654c1023e5f87e",
+      "scripts/run_int_kv_table14_15_matrix.py": "b57b985ada939389206cb0d8137f1701216d609a3d231bfb1892c2b95b7434e7",
+      "scripts/run_kv_calibration_robustness.py": "4af1ca5de75b41ac1dc9b9c8967d13993434de4048080cff192859edbba49471",
+      "scripts/run_kv_granularity_matrix.py": "39f394a99e14d43277ba060daf70357038ce42d7075304b1f09149b68158c3f2",
+      "scripts/run_kv_layer_sensitivity.py": "619a4a422338dd2bf7bf892f1dbab6550e79ac9abb16754c9c27fc19ba6e75ea",
+      "scripts/run_layer_channel_delta_ce.py": "cd28a506b86db4d5c717f934e16328d37932c7a0b785b4c6e8613a6453ee73ef",
+      "scripts/run_lm_eval_timed.py": "acdcfdf1b9ffd040e0fadaa45b1942e3d1b548433f135fbe1c84e2fcad0cb6fa",
+      "scripts/run_loop1_remaining_resumed.py": "56b1f88d04e774e4a383097b11640ab3f91bf214fc7b7daaed0286e789ddcc95",
+      "scripts/run_mixed_precision_finalists.py": "0da37cf7c46b974cc1e0835c6ddf8b7dac31b0e50a00b3afec11501e6379e190",
+      "scripts/run_mixed_precision_kv_matrix.py": "4c3dba3d879add89b8b726dfc667d51e2432f78f8f3f283e5b467aa930ec1043",
+      "scripts/run_momentum_ablation_matrix.py": "8d6469daef62830027c2c937e62ac396cdb186af05f330eea4a5a907ae1c6fd8",
+      "scripts/run_native_nvfp4_method_screen.py": "3f8714578306432462d60e6b73b04905bb68433dd6fa7e4ed916dcb56103c28f",
+      "scripts/run_native_wa_bf16_screen.py": "fb102e51d601a442b61afd582e321e5949eac2cfaa0c07fb7ddda53c10e44cc3",
+      "scripts/run_native_wa_export_queue.py": "6abac94be9ddb192cdd306dbf13398502b6f88ee14047027cc71a8d0befa2de3",
+      "scripts/run_ouro_chrome_trace_queue.py": "549b3ddf9a776b2235d704bcf097e725e9c083860f5bd97748937222a8150fa8",
+      "scripts/run_ouro_remaining_queue.py": "d0e698329cc1e986941a062bab9c49b33d43e6fc069b94036c41918ad42d82fb",
+      "scripts/run_ouro_repeatability_queue.py": "3c44894a3b9ef6feb7d91c24d9360dc214ef1bcf320ccd93717a0b27247de05b",
+      "scripts/run_ouro_rotation_math500.py": "3e36126efec9527380e822aaa25cda4caaa1f98595c67abaa3c759892a34e7fb",
+      "scripts/run_ouro_selected_kv.py": "2186aa2dd115633af04fa6b32f996d0edbf4eca8dfe98dd9f60b0219e0b3f159",
+      "scripts/run_ouro_selective_w4_matrix.py": "59de40bab4307007ea068d1ac9a1390c8fb08156bdd6217f19a286cabd4d0b76",
+      "scripts/run_ouro_selective_w4a_matrix.py": "cd59845df8008f51137ee40093b152375afa138e949ff3ced49c752090b3dad4",
+      "scripts/run_ouro_table5_queue.py": "f102ac7637e4652e10da8e7ddc6c5e68228488fefa024841778bdee184fbfaf7",
+      "scripts/run_ouro_vllm_gsm8k.py": "8bdce3830b7e84376b343fda3bec6be9a1efa60a38527e9607d39c2acb30edb6",
+      "scripts/run_packed_affine_sweep.py": "eb8fd503a5aaf83ed3c236aa0ba580fd67e152ec2689e0d868e9f02a54db9b4a",
+      "scripts/run_presinq_native_screen.py": "77d9e4f4ab52e6c155b87afc20e7c674e145b0637ef7053fac0f42dc54123168",
+      "scripts/run_q08_offset_sweep.py": "595e1b0a8a1752a88f7ab3d5a1f9e2ede337768248ae93aa9ec9d6f6d0082d44",
+      "scripts/run_q08_uniform_bits.py": "990ee1bc1c1b5fc5f56d8b787ace89748eaf24645868d6c014a66e47d0b9b259",
+      "scripts/run_residual_activation_matrix.py": "02a21358423ea5ccb2ec83fc3b047073f641494bb61bb6b62837e648ee0693cf",
+      "scripts/run_residual_kv_matrix.py": "392e763fc90e4ea7e3b664e8f70d009a441c0301e87ac52ae5ec033d79013bfc",
+      "scripts/run_selected_wa_kv_validation.py": "3373b6d950c61a7ae8b5af144e01852cfc519e616e4a77416fa955d1f4e3c3f7",
+      "scripts/run_sinq_matrix.py": "3b87d45fb75b69c59a18a9bcb6efde414089671e1c4d579902719cac86a4a6c5",
+      "scripts/run_table10_calibration_queue.py": "e527825b7356cbf5a5d02e0581a8165b38fee0e93874937b48476bbe1cc8179f",
+      "scripts/run_top3_combined_kv.py": "bebfd392c92c744491502ff7fc59b74bbb51dd4be5cf74cc5146c1b4efb5de56",
+      "scripts/select_heatmap_delta_ce_cells.py": "a16a13f8c20b352759769389fb1690cc45b3243cc49824121ee35e313f4cf794",
+      "scripts/select_kv_codec_full_policies.py": "84c5655c547b3faf6052e998c6791ea41105ecfdf5d7137a53757259a75c1991",
+      "scripts/smoke_ouro_nvfp4_vllm.py": "4d6f608d2b9027a6b5f9071511fdef2aaca37e6cad917ec3862c5cf6cc1b3fa6",
+      "scripts/smoke_ouro_recurrent_vllm.py": "bb159fa91c75084a1ff0ae1fdaf6fd65f29a01e6b49d79a16e06c9aec98df749",
+      "scripts/summarize_channel_profile_costs.py": "1835ebab6097d373c9fcab3ccfc0208c2eb392a6c425d1093bb6f5b704b70998",
+      "scripts/summarize_dynamic_lse.py": "66f32fba2b6031ddc80224194a509ec9eadbaebc06ee7832e06c70acfbb1c6e8",
+      "scripts/summarize_graph_experiments.py": "0877b04fb282eb7afafdfcd7e6453b92e6ffbb3521544e7c0719bd87c748f275",
+      "scripts/summarize_int_scale_matrix.py": "4c937809ee46b2576e1aed96996a5525fd2798bcb9b8f83e65eef3dcf0bccb74",
+      "scripts/summarize_kv_cost_analysis.py": "cd409cfad8ecebf0d8d8a3aac1c829342da463f0a36a3ae7ccf63774fef0c563",
+      "scripts/summarize_ouro_attention_precision.py": "a0bf777fc366cc198629c2c467d0fd1f98335bbcca6ae876bb61fe9c68566319",
+      "scripts/summarize_ouro_chrome_traces.py": "47d7beb592fd8049fb61c3e82a0298b4f65f5c35890342f2cc97f73a56843632",
+      "scripts/summarize_ouro_fair_comparison.py": "d16891bdbbb74901cebda757b426e524e447c20bee7c87b36b8ab00bae5f791a",
+      "scripts/summarize_ouro_final_optimization.py": "5222fc60b20210f6786a66005dd32ff3f6a750e5bc92f67654ec325bf0d8f7c7",
+      "scripts/summarize_quantization_goal.py": "73dd960173630d82664d1cd240f71cb329d15fa9842674cd9ca1df115e1370c6",
+      "scripts/summarize_recurrent_full.py": "519bff759541a33cceb3c3513a9537d9272b9b0e8b9c58e3b9f6267bf8b0979a",
+      "scripts/summarize_recurrent_mixed.py": "406b963dbb92587f781402a6b386bdf55d98bf053aab4c0f25432c931510662c",
+      "scripts/summarize_recurrent_screen.py": "68fcc410a93a0d667b97e7ab2a334c55eaa1d7250a8d3ad8f245ce8988db7f9d",
+      "scripts/summarize_recurrent_xqa.py": "c4b3718a49fd016a7d1b7ef3d1772cd0f31ef038f5a596ea0eb63460e035bc62",
+      "scripts/summarize_runpod_benchmarks.py": "9d18c35c572aed7b4ca1ba8f08bd79db2094bcee61252733f7c3c1a19189cf5f",
+      "scripts/summarize_scan_full.py": "7c10989917660da80a64ae4414dd12bec35d9d8e3d24a0b0d6670081bc3fc234",
+      "scripts/summarize_xqa_exploration.py": "9abeef642993d443c99a5a98970c89688326c45a7245c0cad617442506664f4a",
+      "scripts/trace_channel_capacity_schedule.py": "6fdb5972552f69c96318764f1e50fda3d08487db3b471c7d56f8ebaafec905ba",
+      "scripts/validate_native_kv.py": "cb726f0a16755d968317cb133bf232de1b0135dfbc0fa2f73efab2557af5775f",
+      "scripts/validate_nvfp4_modelopt_hopper.py": "079cf32815542feb97b0784696bbac1488c57d562c70905cb67355e74160e8f8",
+      "scripts/validate_pre_sinq_checkpoint.py": "5f6a7d4daed934b88e30eb4dde79f62a272782b69a35a220f4899be966af8a2d",
+      "scripts/validate_reqat_e1m2_modelopt.py": "31a076de14ab0bd99facfa7a2ca19eac2aa947c2872fe64c70f08e87f40b0e0c",
+      "plugins/ouro_quant_vllm/pyproject.toml": "c16833fa241d6219d645f79979c5df2bc4a8c3494a74a35d4c8ac38d59e96fcc",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/__init__.py": "09961d7d247f2e2667eaa45747270f53172d32422e3a09391a3ad65d149822f8",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/channel_requests.py": "3389cf929bedb1dd3d4d38a7dec8ab598cba3fa05fef8746f3d32d9fa34c2474",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/fused_norm.py": "ea5b02eda1eed9280729ac2801ecefb8fc41fd82002bd76bf9ecce5c7c586e54",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/fused_rope_cache.py": "7765006ecd10e9b8562f5d479cff12f4471c807bab271954a699bb381dc97c83",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/kv_rotation.py": "b7d4ff4342d42a07d7791180c6ffad931a8d283ab0b46c7b1e4ae32865c0d436",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/loopq_bridge.py": "be35be25a58efdc2fb0ebc7f7101b01d9c8b0aebd66b0f0a60c21fe5557cf468",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/model.py": "806182c39a5693efd7895a607be17c44ccbc2b74d4e6624d90596c7a4cfb1f0e",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/native_attention.py": "f2ddc785386102a38a40cc8b108ab2c4bfb7fe99cd3dcbe2b400f747b59483aa",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/nvfp4_qdq.py": "b55c0270fdb6131848456141261d1c6b984f3f88ed31ac561ee7f981cb6a593c",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_affine_attention.py": "0a9a0a1018341df5261254fdc9fabf1e1534381d91ab7b91a8a7c8ef6e0bccc2",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_rotated_attention.py": "d8f7b9ec10f0c1c15e6043fc26444b4a6a376bfffdbcd8232e0ce577e270a01c",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/packed_rotated_table1_attention.py": "7ae22f08c2efd240fb4c549943eaaae972a3161f482b91cbb5194bda3fd6b1a7",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/recurrent_attention.py": "2e6ff15630c0757a9a36ca14818535f7a605be2800de8083841493ba25524dc3",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/residual_activation.py": "8cca2e8ad00594220b2658c4be851efdc9627506c01900e8647c89463b8bc877",
+      "plugins/ouro_quant_vllm/src/ouro_quant_vllm/residual_kv.py": "4cad104ba295e294f87eb27eada1d968be46a7d5b4b0c08edb52e8bbebfc8e49",
+      "plugins/huginn_vllm_validated/pyproject.toml": "4538f596380c7d8487189accd14597cb2ed237dd76bb719a7057a95abefb4f14",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/__init__.py": "5e18f53c2fdb7911b6cc921decc552d1fa92369fc115e1a1199288e58bcb16cc",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/attention_bank.py": "8482b5175982303d7d999a5e0fdf71e2e6f09d5f34a88e57a6bdeb3df58a7dc8",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/grouped_attention.py": "1e7df3c0c3f7cebc40cb977c11926a0b2c032653c20454a160a7ecbc21fd7b63",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/loopq_bridge.py": "5cb3b60e24a2c683976d6229dd89135a04e3167ede696742a55a5dd327d49f7e",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/model.py": "252b5b87d06ee17896fc91945a645c1e528290c70be7e2129bf9f2271177e383",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/native_attention.py": "93843837ccd9b3393a080391d9c013a9586304af1acc1d850e52e8fbec7d2294",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/runtime.py": "c9b71f1f48d75475ea299eab01d5bcf6f38445a7c7eec62fe2d5205b47d6c218",
+      "plugins/huginn_vllm_validated/src/huginn_vllm_validated/wavefront.py": "a74f9b084a9e2708c1211dc9fb6cb20de23d61e17868e0967fcf0e6cd27eb125"
+    },
+    "scope": "repository source at launch; installed dependencies are identified separately by environment metadata",
+    "changed_during_run": [],
+    "source_tree_unchanged_at_exit": true
+  },
+  "result_path": "/workspace/loopq_data/curve_a4_s0/eval_final/lambada_openai/harness/ByteDance__Ouro-1.4B/results_2026-09-16T23-39-38.030831.json",
+  "samples": {
+    "lambada_openai": {
+      "original": 5153,
+      "effective": 5153
+    }
+  },
+  "metrics": {
+    "lambada_openai": {
+      "name": "lambada_openai",
+      "alias": "lambada_openai",
+      "sample_len": 5153,
+      "perplexity,none": 8.383131365672106,
+      "perplexity_stderr,none": 0.23103998744055027,
+      "acc,none": 0.5604502231709684,
+      "acc_stderr,none": 0.006914879684264954
+    }
+  },
+  "task_configs": {
+    "lambada_openai": {
+      "task": "lambada_openai",
+      "dataset_path": "EleutherAI/lambada_openai",
+      "dataset_name": "default",
+      "test_split": "test",
+      "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
+      "doc_to_target": "{{' '+text.split(' ')[-1]}}",
+      "unsafe_code": false,
+      "description": "",
+      "target_delimiter": " ",
+      "fewshot_delimiter": "\n\n",
+      "fewshot_config": {
+        "sampler": "default",
+        "split": null,
+        "process_docs": null,
+        "fewshot_indices": null,
+        "samples": null,
+        "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
+        "doc_to_choice": null,
+        "doc_to_target": "{{' '+text.split(' ')[-1]}}",
+        "gen_prefix": null,
+        "fewshot_delimiter": "\n\n",
+        "target_delimiter": " "
+      },
+      "num_fewshot": 0,
+      "metric_list": [
+        {
+          "metric": "perplexity",
+          "aggregation": "perplexity",
+          "higher_is_better": false
+        },
+        {
+          "metric": "acc",
+          "aggregation": "mean",
+          "higher_is_better": true
+        }
+      ],
+      "output_type": "loglikelihood",
+      "repeats": 1,
+      "should_decontaminate": true,
+      "doc_to_decontamination_query": "{{text}}",
+      "metadata": {
+        "version": 1.0,
+        "pretrained": "ByteDance/Ouro-1.4B",
+        "revision": "e3b1e0993b1231a51d6069a870476dda4162c00a",
+        "trust_remote_code": true,
+        "dtype": "bfloat16",
+        "tensor_parallel_size": 1,
+        "max_model_len": 4096,
+        "gpu_memory_utilization": 0.9,
+        "enforce_eager": true,
+        "max_num_seqs": 32,
+        "max_num_batched_tokens": 4096,
+        "enable_prefix_caching": false,
+        "seed": 0,
+        "config_source": "/root/venv-LT/lib/python3.11/site-packages/lm_eval/tasks/lambada/lambada_openai.yaml"
+      }
+    }
+  }
+}

loopq_quantization/results/loopq_w4a4_completed/lambada_openai/timing.json

@@ -0,0 +1,9 @@
+{
+  "model_init_seconds": 33.6115588368848,
+  "generate_until_seconds": 0.0,
+  "generated_samples": 0,
+  "generated_tokens": 0,
+  "generation_calls": 0,
+  "completed": true,
+  "process_seconds": 245.44895535707474
+}

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/README.md

@@ -0,0 +1,76 @@
+# Task-name
+
+### Paper
+
+Title: `Measuring Massive Multitask Language Understanding`
+
+Abstract: `https://arxiv.org/abs/2009.03300`
+
+`The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more.`
+
+Homepage: `https://github.com/hendrycks/test`
+
+Note: The `Flan` variants are derived from [here](https://github.com/jasonwei20/flan-2), and as described in Appendix D.1 of [Scaling Instruction-Finetuned Language Models](https://arxiv.org/abs/2210.11416).
+
+### Citation
+
+```
+@article{hendryckstest2021,
+  title={Measuring Massive Multitask Language Understanding},
+  author={Dan Hendrycks and Collin Burns and Steven Basart and Andy Zou and Mantas Mazeika and Dawn Song and Jacob Steinhardt},
+  journal={Proceedings of the International Conference on Learning Representations (ICLR)},
+  year={2021}
+}
+
+@article{hendrycks2021ethics,
+  title={Aligning AI With Shared Human Values},
+  author={Dan Hendrycks and Collin Burns and Steven Basart and Andrew Critch and Jerry Li and Dawn Song and Jacob Steinhardt},
+  journal={Proceedings of the International Conference on Learning Representations (ICLR)},
+  year={2021}
+}
+```
+
+### Groups, Tags, and Tasks
+
+#### Groups
+
+* `mmlu`: `Original multiple-choice MMLU benchmark`
+* `mmlu_continuation`: `MMLU but with continuation prompts`
+* `mmlu_generative`: `MMLU generation`
+
+MMLU is the original benchmark as implemented by Hendrycks et al. with the choices in context and the answer letters (e.g `A`, `B`, `C`, `D`) in the continuation.
+`mmlu_continuation` is a cloze-style variant without the choices in context and the full answer choice in the continuation.
+`mmlu_generative` is a generation variant, similar to the original but the LLM is asked to generate the correct answer letter.
+
+
+#### Subgroups
+
+* `mmlu_stem'
+* `mmlu_humanities'
+* `mmlu_social_sciences'
+* `mmlu_other'
+
+Subgroup variants are prefixed with the subgroup name, e.g. `mmlu_stem_continuation`.
+
+### Checklist
+
+For adding novel benchmarks/datasets to the library:
+* [x] Is the task an existing benchmark in the literature?
+  * [x] Have you referenced the original paper that introduced the task?
+  * [x] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
+
+
+If other tasks on this dataset are already supported:
+* [x] Is the "Main" variant of this task clearly denoted?
+* [x] Have you provided a short sentence in a README on what each new variant adds / evaluates?
+* [x] Have you noted which, if any, published evaluation setups are matched by this variant?
+
+# changelog
+ver 1: PR #497
+switch to original implementation
+
+ver 2: PR #2116
+add missing newline in description.
+
+PR #3137
+Fix `mmlu_continuation` subgroup names to fit other variants, and switch dataset from `hails/mmlu_no_train` to `cais/mmlu` in all subtasks.

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/_generate_configs.py

@@ -0,0 +1,159 @@
+# noqa
+"""
+Take in a YAML, and output all "other" splits with this YAML
+"""
+
+import argparse
+import logging
+import os
+
+import yaml
+from tqdm import tqdm
+
+
+eval_logger = logging.getLogger(__name__)
+
+
+SUBJECTS = {
+    "abstract_algebra": "stem",
+    "anatomy": "stem",
+    "astronomy": "stem",
+    "business_ethics": "other",
+    "clinical_knowledge": "other",
+    "college_biology": "stem",
+    "college_chemistry": "stem",
+    "college_computer_science": "stem",
+    "college_mathematics": "stem",
+    "college_medicine": "other",
+    "college_physics": "stem",
+    "computer_security": "stem",
+    "conceptual_physics": "stem",
+    "econometrics": "social_sciences",
+    "electrical_engineering": "stem",
+    "elementary_mathematics": "stem",
+    "formal_logic": "humanities",
+    "global_facts": "other",
+    "high_school_biology": "stem",
+    "high_school_chemistry": "stem",
+    "high_school_computer_science": "stem",
+    "high_school_european_history": "humanities",
+    "high_school_geography": "social_sciences",
+    "high_school_government_and_politics": "social_sciences",
+    "high_school_macroeconomics": "social_sciences",
+    "high_school_mathematics": "stem",
+    "high_school_microeconomics": "social_sciences",
+    "high_school_physics": "stem",
+    "high_school_psychology": "social_sciences",
+    "high_school_statistics": "stem",
+    "high_school_us_history": "humanities",
+    "high_school_world_history": "humanities",
+    "human_aging": "other",
+    "human_sexuality": "social_sciences",
+    "international_law": "humanities",
+    "jurisprudence": "humanities",
+    "logical_fallacies": "humanities",
+    "machine_learning": "stem",
+    "management": "other",
+    "marketing": "other",
+    "medical_genetics": "other",
+    "miscellaneous": "other",
+    "moral_disputes": "humanities",
+    "moral_scenarios": "humanities",
+    "nutrition": "other",
+    "philosophy": "humanities",
+    "prehistory": "humanities",
+    "professional_accounting": "other",
+    "professional_law": "humanities",
+    "professional_medicine": "other",
+    "professional_psychology": "social_sciences",
+    "public_relations": "social_sciences",
+    "security_studies": "social_sciences",
+    "sociology": "social_sciences",
+    "us_foreign_policy": "social_sciences",
+    "virology": "other",
+    "world_religions": "humanities",
+}
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--base_yaml_path", required=True)
+    parser.add_argument("--save_prefix_path", default="mmlu")
+    parser.add_argument("--cot_prompt_path", default=None)
+    parser.add_argument("--task_prefix", default="")
+    parser.add_argument("--group_prefix", default="")
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+
+    # get filename of base_yaml so we can `"include": ` it in our "other" YAMLs.
+    base_yaml_name = os.path.split(args.base_yaml_path)[-1]
+    with open(args.base_yaml_path, encoding="utf-8") as f:
+        base_yaml = yaml.full_load(f)
+
+    if args.cot_prompt_path is not None:
+        import json
+
+        with open(args.cot_prompt_path, encoding="utf-8") as f:
+            cot_file = json.load(f)
+
+    ALL_CATEGORIES = []
+    for subject, category in tqdm(SUBJECTS.items()):
+        if category not in ALL_CATEGORIES:
+            ALL_CATEGORIES.append(category)
+
+        if args.cot_prompt_path is not None:
+            description = cot_file[subject]
+        else:
+            description = f"The following are multiple choice questions (with answers) about {' '.join(subject.split('_'))}.\n\n"
+
+        yaml_dict = {
+            "include": base_yaml_name,
+            "tag": f"mmlu_{args.task_prefix}_{category}"
+            if args.task_prefix != ""
+            else f"mmlu_{category}",
+            "task": f"mmlu_{args.task_prefix}_{subject}"
+            if args.task_prefix != ""
+            else f"mmlu_{subject}",
+            "task_alias": subject.replace("_", " "),
+            "dataset_name": subject,
+            "description": description,
+        }
+
+        file_save_path = args.save_prefix_path + f"_{subject}.yaml"
+        eval_logger.info(f"Saving yaml for subset {subject} to {file_save_path}")
+        with open(file_save_path, "w", encoding="utf-8") as yaml_file:
+            yaml.dump(
+                yaml_dict,
+                yaml_file,
+                allow_unicode=True,
+                default_style='"',
+            )
+
+    if args.task_prefix != "":
+        mmlu_subcategories = [
+            f"mmlu_{args.task_prefix}_{category}" for category in ALL_CATEGORIES
+        ]
+    else:
+        mmlu_subcategories = [f"mmlu_{category}" for category in ALL_CATEGORIES]
+
+    if args.group_prefix != "":
+        file_save_path = args.group_prefix + ".yaml"
+    else:
+        file_save_path = args.save_prefix_path + ".yaml"
+
+    eval_logger.info(f"Saving benchmark config to {file_save_path}")
+    with open(file_save_path, "w", encoding="utf-8") as yaml_file:
+        yaml.dump(
+            {
+                "group": f"mmlu_{args.task_prefix}"
+                if args.task_prefix != ""
+                else "mmlu",
+                "task": mmlu_subcategories,
+            },
+            yaml_file,
+            indent=4,
+            default_flow_style=False,
+        )

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/_continuation_template_yaml

@@ -0,0 +1,11 @@
+dataset_path: cais/mmlu
+output_type: multiple_choice
+test_split: test
+fewshot_split: dev
+fewshot_config:
+  sampler: first_n
+doc_to_text: "Question: {{question.strip()}}\nAnswer:"
+doc_to_choice: "{{choices}}"
+doc_to_target: "{{answer}}"
+metadata:
+  version: 1.0

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/_mmlu.yaml

@@ -0,0 +1,32 @@
+group: mmlu_continuation
+group_alias: mmlu (continuation)
+task:
+  - group: stem
+    task:
+      - mmlu_stem_continuation
+    aggregate_metric_list:
+      - metric: acc
+        weight_by_size: True
+  - group: other
+    task:
+      - mmlu_other_continuation
+    aggregate_metric_list:
+      - metric: acc
+        weight_by_size: True
+  - group: social sciences
+    task:
+      - mmlu_social_sciences_continuation
+    aggregate_metric_list:
+      - metric: acc
+        weight_by_size: True
+  - group: humanities
+    task:
+      - mmlu_humanities_continuation
+    aggregate_metric_list:
+      - metric: acc
+        weight_by_size: True
+aggregate_metric_list:
+  - metric: acc
+    weight_by_size: True
+metadata:
+  version: 2

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_abstract_algebra.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "abstract_algebra"
+"description": "The following are questions (with answers) about abstract\
+  \ algebra.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_abstract_algebra_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_anatomy.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "anatomy"
+"description": "The following are questions (with answers) about anatomy.\n\
+  \n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_anatomy_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_astronomy.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "astronomy"
+"description": "The following are questions (with answers) about astronomy.\n\
+  \n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_astronomy_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_business_ethics.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "business_ethics"
+"description": "The following are questions (with answers) about business\
+  \ ethics.\n\n"
+"tag": "mmlu_other_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_business_ethics_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_clinical_knowledge.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "clinical_knowledge"
+"description": "The following are questions (with answers) about clinical\
+  \ knowledge.\n\n"
+"tag": "mmlu_other_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_clinical_knowledge_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_college_biology.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "college_biology"
+"description": "The following are questions (with answers) about college\
+  \ biology.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_college_biology_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_college_chemistry.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "college_chemistry"
+"description": "The following are questions (with answers) about college\
+  \ chemistry.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_college_chemistry_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_college_computer_science.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "college_computer_science"
+"description": "The following are questions (with answers) about college\
+  \ computer science.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_college_computer_science_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_college_mathematics.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "college_mathematics"
+"description": "The following are questions (with answers) about college\
+  \ mathematics.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_college_mathematics_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_college_medicine.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "college_medicine"
+"description": "The following are questions (with answers) about college\
+  \ medicine.\n\n"
+"tag": "mmlu_other_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_college_medicine_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_college_physics.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "college_physics"
+"description": "The following are questions (with answers) about college\
+  \ physics.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_college_physics_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_computer_security.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "computer_security"
+"description": "The following are questions (with answers) about computer\
+  \ security.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_computer_security_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_conceptual_physics.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "conceptual_physics"
+"description": "The following are questions (with answers) about conceptual\
+  \ physics.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_conceptual_physics_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_econometrics.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "econometrics"
+"description": "The following are questions (with answers) about econometrics.\n\
+  \n"
+"tag": "mmlu_social_sciences_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_econometrics_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_electrical_engineering.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "electrical_engineering"
+"description": "The following are questions (with answers) about electrical\
+  \ engineering.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_electrical_engineering_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_elementary_mathematics.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "elementary_mathematics"
+"description": "The following are questions (with answers) about elementary\
+  \ mathematics.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_elementary_mathematics_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_formal_logic.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "formal_logic"
+"description": "The following are questions (with answers) about formal\
+  \ logic.\n\n"
+"tag": "mmlu_humanities_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_formal_logic_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_global_facts.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "global_facts"
+"description": "The following are questions (with answers) about global\
+  \ facts.\n\n"
+"tag": "mmlu_other_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_global_facts_continuation"

loopq_quantization/results/loopq_w4a4_completed/mmlu/installed_task_snapshot/continuation/mmlu_high_school_biology.yaml

@@ -0,0 +1,6 @@
+"dataset_name": "high_school_biology"
+"description": "The following are questions (with answers) about high\
+  \ school biology.\n\n"
+"tag": "mmlu_stem_continuation"
+"include": "_continuation_template_yaml"
+"task": "mmlu_high_school_biology_continuation"