| { | |
| "conditions": { | |
| "candidate": { | |
| "n": 156, | |
| "accuracy": 0.9551282051282052, | |
| "balanced_accuracy": 0.9672897196261683, | |
| "by_category": { | |
| "abstain": 1.0, | |
| "answer": 0.9345794392523364 | |
| }, | |
| "leaks": 0, | |
| "mean_tokens": 22.666666666666668 | |
| }, | |
| "hf_base": { | |
| "n": 156, | |
| "accuracy": 0.48717948717948717, | |
| "balanced_accuracy": 0.6095746709898913, | |
| "by_category": { | |
| "abstain": 0.9387755102040817, | |
| "answer": 0.2803738317757009 | |
| }, | |
| "leaks": 0, | |
| "mean_tokens": 38.73717948717949 | |
| }, | |
| "native_base": { | |
| "n": 156, | |
| "accuracy": 0.48717948717948717, | |
| "balanced_accuracy": 0.6095746709898913, | |
| "by_category": { | |
| "abstain": 0.9387755102040817, | |
| "answer": 0.2803738317757009 | |
| }, | |
| "leaks": 0, | |
| "mean_tokens": 39.18589743589744 | |
| }, | |
| "native_challenger": { | |
| "n": 156, | |
| "accuracy": 0.6602564102564102, | |
| "balanced_accuracy": 0.7523364485981308, | |
| "by_category": { | |
| "abstain": 1.0, | |
| "answer": 0.5046728971962616 | |
| }, | |
| "leaks": 0, | |
| "mean_tokens": 40.967948717948715 | |
| }, | |
| "cpu_trained": { | |
| "n": 156, | |
| "accuracy": 0.5961538461538461, | |
| "balanced_accuracy": 0.6834827388899485, | |
| "by_category": { | |
| "abstain": 0.9183673469387755, | |
| "answer": 0.4485981308411215 | |
| }, | |
| "leaks": 0, | |
| "mean_tokens": 0.0 | |
| }, | |
| "lexical_rules": { | |
| "n": 156, | |
| "accuracy": 0.2692307692307692, | |
| "balanced_accuracy": 0.19626168224299065, | |
| "by_category": { | |
| "abstain": 0.0, | |
| "answer": 0.3925233644859813 | |
| }, | |
| "leaks": 0, | |
| "mean_tokens": 0.0 | |
| }, | |
| "gpt_oss_20b": { | |
| "n": 156, | |
| "accuracy": 0.6153846153846154, | |
| "balanced_accuracy": 0.7140949837879077, | |
| "by_category": { | |
| "abstain": 0.9795918367346939, | |
| "answer": 0.4485981308411215 | |
| }, | |
| "leaks": 0, | |
| "mean_tokens": 58.166666666666664 | |
| } | |
| }, | |
| "comparisons": { | |
| "hf_base": { | |
| "groups": 103, | |
| "delta": 0.5184466019417475, | |
| "interval95": [ | |
| 0.4385113268608415, | |
| 0.6019579288025887 | |
| ] | |
| }, | |
| "native_base": { | |
| "groups": 103, | |
| "delta": 0.5275080906148868, | |
| "interval95": [ | |
| 0.4482200647249191, | |
| 0.6100323624595468 | |
| ] | |
| }, | |
| "native_challenger": { | |
| "groups": 103, | |
| "delta": 0.34012944983818777, | |
| "interval95": [ | |
| 0.2576051779935275, | |
| 0.42556634304207114 | |
| ] | |
| }, | |
| "cpu_trained": { | |
| "groups": 103, | |
| "delta": 0.36391585760517803, | |
| "interval95": [ | |
| 0.2823624595469255, | |
| 0.44482605177993517 | |
| ] | |
| }, | |
| "lexical_rules": { | |
| "groups": 103, | |
| "delta": 0.6181229773462783, | |
| "interval95": [ | |
| 0.5307443365695793, | |
| 0.7022653721682848 | |
| ] | |
| }, | |
| "gpt_oss_20b": { | |
| "groups": 103, | |
| "delta": 0.38220064724919095, | |
| "interval95": [ | |
| 0.29563106796116506, | |
| 0.47120954692556616 | |
| ] | |
| } | |
| }, | |
| "release_gate_passed": true, | |
| "superior": true, | |
| "minimum_quality": true, | |
| "all_prompt_hashes_exact": true, | |
| "fp32_reload": { | |
| "tensors": 208, | |
| "all_saved_tensors_exact": true, | |
| "storage": "float32", | |
| "weights_sha256": "075107999fddb3b2a28f424beac0a329199aa7f6689ed93df3ff14dd674d0a2d" | |
| }, | |
| "conversion": { | |
| "source_weights_sha256": "532513c9c29daa15fe0040d31db93fb26a82ad92aa4626e6947ed915e0bb878a", | |
| "converted_weights_sha256": "075107999fddb3b2a28f424beac0a329199aa7f6689ed93df3ff14dd674d0a2d", | |
| "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", | |
| "source_tensors": 304, | |
| "converted_tensors": 208, | |
| "qkv_groups": 24, | |
| "qkv_rank": 48, | |
| "scaling_preserved": true, | |
| "all_source_tensors_consumed": true, | |
| "max_float32_update_error": 1.862645149230957e-09, | |
| "conversion": "A concatenation, B block diagonal, rank/alpha tripled; output projection -> lm_head", | |
| "mapping": { | |
| "base_model.model.model.layers.0.linear_attn": "base_model.model.model.language_model.layers.0.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.1.linear_attn": "base_model.model.model.language_model.layers.1.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.10.linear_attn": "base_model.model.model.language_model.layers.10.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.12.linear_attn": "base_model.model.model.language_model.layers.12.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.13.linear_attn": "base_model.model.model.language_model.layers.13.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.14.linear_attn": "base_model.model.model.language_model.layers.14.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.16.linear_attn": "base_model.model.model.language_model.layers.16.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.17.linear_attn": "base_model.model.model.language_model.layers.17.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.18.linear_attn": "base_model.model.model.language_model.layers.18.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.2.linear_attn": "base_model.model.model.language_model.layers.2.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.20.linear_attn": "base_model.model.model.language_model.layers.20.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.21.linear_attn": "base_model.model.model.language_model.layers.21.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.22.linear_attn": "base_model.model.model.language_model.layers.22.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.24.linear_attn": "base_model.model.model.language_model.layers.24.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.25.linear_attn": "base_model.model.model.language_model.layers.25.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.26.linear_attn": "base_model.model.model.language_model.layers.26.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.28.linear_attn": "base_model.model.model.language_model.layers.28.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.29.linear_attn": "base_model.model.model.language_model.layers.29.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.30.linear_attn": "base_model.model.model.language_model.layers.30.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.4.linear_attn": "base_model.model.model.language_model.layers.4.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.5.linear_attn": "base_model.model.model.language_model.layers.5.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.6.linear_attn": "base_model.model.model.language_model.layers.6.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.8.linear_attn": "base_model.model.model.language_model.layers.8.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.9.linear_attn": "base_model.model.model.language_model.layers.9.linear_attn.in_proj_qkv", | |
| "base_model.model.model.layers.0.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.0.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.0.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.0.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.0.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.0.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.0.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.0.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.1.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.1.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.1.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.1.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.1.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.1.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.1.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.1.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.2.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.2.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.2.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.2.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.2.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.2.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.2.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.2.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.3.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.3.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.3.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.3.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.3.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.3.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.3.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.3.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.3.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.3.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.3.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.3.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.3.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.3.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.3.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.3.self_attn.o_proj.lora_B.weight", | |
| "base_model.model.model.layers.4.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.4.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.4.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.4.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.4.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.4.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.4.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.4.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.5.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.5.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.5.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.5.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.5.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.5.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.5.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.5.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.6.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.6.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.6.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.6.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.6.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.6.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.6.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.6.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.7.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.7.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.7.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.7.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.7.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.7.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.7.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.7.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.7.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.7.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.7.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.7.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.7.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.7.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.7.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.7.self_attn.o_proj.lora_B.weight", | |
| "base_model.model.model.layers.8.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.8.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.8.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.8.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.8.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.8.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.8.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.8.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.9.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.9.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.9.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.9.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.9.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.9.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.9.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.9.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.10.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.10.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.10.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.10.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.10.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.10.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.10.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.10.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.11.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.11.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.11.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.11.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.11.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.11.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.11.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.11.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.11.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.11.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.11.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.11.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.11.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.11.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.11.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.11.self_attn.o_proj.lora_B.weight", | |
| "base_model.model.model.layers.12.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.12.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.12.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.12.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.12.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.12.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.12.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.12.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.13.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.13.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.13.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.13.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.13.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.13.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.13.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.13.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.14.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.14.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.14.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.14.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.14.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.14.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.14.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.14.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.15.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.15.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.15.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.15.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.15.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.15.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.15.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.15.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.15.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.15.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.15.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.15.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.15.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.15.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.15.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.15.self_attn.o_proj.lora_B.weight", | |
| "base_model.model.model.layers.16.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.16.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.16.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.16.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.16.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.16.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.16.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.16.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.17.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.17.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.17.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.17.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.17.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.17.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.17.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.17.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.18.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.18.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.18.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.18.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.18.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.18.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.18.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.18.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.19.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.19.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.19.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.19.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.19.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.19.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.19.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.19.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.19.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.19.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.19.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.19.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.19.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.19.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.19.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.19.self_attn.o_proj.lora_B.weight", | |
| "base_model.model.model.layers.20.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.20.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.20.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.20.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.20.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.20.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.20.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.20.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.21.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.21.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.21.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.21.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.21.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.21.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.21.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.21.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.22.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.22.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.22.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.22.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.22.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.22.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.22.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.22.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.23.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.23.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.23.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.23.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.23.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.23.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.23.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.23.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.23.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.23.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.23.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.23.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.23.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.23.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.23.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.23.self_attn.o_proj.lora_B.weight", | |
| "base_model.model.model.layers.24.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.24.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.24.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.24.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.24.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.24.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.24.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.24.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.25.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.25.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.25.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.25.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.25.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.25.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.25.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.25.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.26.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.26.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.26.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.26.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.26.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.26.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.26.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.26.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.27.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.27.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.27.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.27.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.27.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.27.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.27.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.27.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.27.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.27.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.27.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.27.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.27.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.27.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.27.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.27.self_attn.o_proj.lora_B.weight", | |
| "base_model.model.model.layers.28.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.28.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.28.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.28.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.28.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.28.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.28.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.28.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.29.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.29.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.29.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.29.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.29.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.29.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.29.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.29.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.30.linear_attn.in_proj_z.lora_A.weight": "base_model.model.model.language_model.layers.30.linear_attn.in_proj_z.lora_A.weight", | |
| "base_model.model.model.layers.30.linear_attn.in_proj_z.lora_B.weight": "base_model.model.model.language_model.layers.30.linear_attn.in_proj_z.lora_B.weight", | |
| "base_model.model.model.layers.30.linear_attn.out_proj.lora_A.weight": "base_model.model.model.language_model.layers.30.linear_attn.out_proj.lora_A.weight", | |
| "base_model.model.model.layers.30.linear_attn.out_proj.lora_B.weight": "base_model.model.model.language_model.layers.30.linear_attn.out_proj.lora_B.weight", | |
| "base_model.model.model.layers.31.self_attn.q_proj.lora_A.weight": "base_model.model.model.language_model.layers.31.self_attn.q_proj.lora_A.weight", | |
| "base_model.model.model.layers.31.self_attn.q_proj.lora_B.weight": "base_model.model.model.language_model.layers.31.self_attn.q_proj.lora_B.weight", | |
| "base_model.model.model.layers.31.self_attn.k_proj.lora_A.weight": "base_model.model.model.language_model.layers.31.self_attn.k_proj.lora_A.weight", | |
| "base_model.model.model.layers.31.self_attn.k_proj.lora_B.weight": "base_model.model.model.language_model.layers.31.self_attn.k_proj.lora_B.weight", | |
| "base_model.model.model.layers.31.self_attn.v_proj.lora_A.weight": "base_model.model.model.language_model.layers.31.self_attn.v_proj.lora_A.weight", | |
| "base_model.model.model.layers.31.self_attn.v_proj.lora_B.weight": "base_model.model.model.language_model.layers.31.self_attn.v_proj.lora_B.weight", | |
| "base_model.model.model.layers.31.self_attn.o_proj.lora_A.weight": "base_model.model.model.language_model.layers.31.self_attn.o_proj.lora_A.weight", | |
| "base_model.model.model.layers.31.self_attn.o_proj.lora_B.weight": "base_model.model.model.language_model.layers.31.self_attn.o_proj.lora_B.weight" | |
| }, | |
| "external_load_verified": true, | |
| "verification_report": "evaluation.json" | |
| }, | |
| "base": "Qwen/Qwen3.5-4B", | |
| "base_revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a", | |
| "rows_sha256": "b3c542237dba50153d88d9358db3c05351bdfbd99fa6b7b87362b9751de5bde3", | |
| "seconds_per_request": { | |
| "base": 0.8669059229102388, | |
| "candidate": 0.5922850079359472 | |
| }, | |
| "native_candidate_agreement": 1.0, | |
| "environment": { | |
| "torch": "2.6.0+cu126", | |
| "transformers": "5.22.1", | |
| "peft": "0.21.2" | |
| }, | |
| "gpu": "NVIDIA A100-SXM4-80GB", | |
| "batch_size": 8, | |
| "scope": "Exact reference passage-set agreement, not full-answer accuracy or independent human entailment. Paper-disjoint synthetic teacher annotations; strengthened CPU baseline added after native test result. Final checkpoint fixed before native test, no retraining.", | |
| "original_backbone_config": { | |
| "provider": "tinker", | |
| "purpose": "evidence-original-base", | |
| "reserved_usd": 0.7, | |
| "status": "reserved", | |
| "model": "openai/gpt-oss-20b", | |
| "revision": "6cee5e81ee83917806bbde320786a8fb61efebee", | |
| "max_tokens": 4096, | |
| "reasoning": "low", | |
| "added_after_native_result": true, | |
| "rows_sha256": "b3c542237dba50153d88d9358db3c05351bdfbd99fa6b7b87362b9751de5bde3", | |
| "rates": { | |
| "prefill": 0.18, | |
| "sample": 0.45 | |
| }, | |
| "scope": "Untuned original paper-agent backbone, same narrowed ID-selection task; no training or test-based candidate changes." | |
| } | |
| } | |