| { |
| "checkpoint_model_sha256": "30dac8e68ccf869d872beb382cebd0be8b168de761dfb38f1358d260d4b8d09d", |
| "official_openai_revision": "599476783c6f88508dab8577808b5ead5cbee8d2", |
| "triton_kernels_revision": "9e1e203f64752cf99abf0e44286231c5d5df7e76", |
| "device": "NVIDIA GeForce RTX 5060 Ti", |
| "device_capability": [ |
| 12, |
| 0 |
| ], |
| "torch_version": "2.8.0+cu128", |
| "triton_version": "3.4.0", |
| "reproduction_command": "CUDA_VISIBLE_DEVICES=<gpu> PYTHONPATH=third_party/openai-gpt-oss:third_party/triton/python/triton_kernels python tools/generate_gptoss_mxfp4_reference.py --checkpoint original --tokenizer hf --output-dir rebuilt/mxfp4_reference_outputs --openai-repo third_party/openai-gpt-oss --triton-repo third_party/triton --prompt 'Once upon a time,' --max-new-tokens 64 --context 256 --device cuda:0", |
| "prompt": "Once upon a time,", |
| "prompt_token_ids": [ |
| 430, |
| 446, |
| 258, |
| 395, |
| 11 |
| ], |
| "logits_shape": [ |
| 1, |
| 5, |
| 1024 |
| ], |
| "output_provenance": { |
| "full_logits": "gpt_oss.triton.model.Transformer full-sequence forward", |
| "cached_logits": "gpt_oss.triton.model.Transformer with per-layer Cache", |
| "greedy_continuation": "gpt_oss.triton.model.Transformer with per-layer Cache", |
| "routing": "gpt_oss.torch.model.Transformer MoE equations", |
| "expert_projection": "gpt_oss.torch.model.Transformer MoE equations" |
| }, |
| "full_vs_cached": { |
| "max_abs_diff": 0.0, |
| "mean_abs_diff": 0.0, |
| "all_token_argmax_equal": true |
| }, |
| "triton_full_vs_torch_full": { |
| "max_abs_diff": 0.4375, |
| "mean_abs_diff": 0.0456787645816803, |
| "all_token_argmax_equal": true |
| }, |
| "routing": { |
| "layers": 6, |
| "tokens": 5, |
| "experts_per_token": 4, |
| "includes_router_logits": true |
| }, |
| "expert_projection": { |
| "selected_output_shape_per_layer": [ |
| 5, |
| 4, |
| 128 |
| ], |
| "includes_weighted_sum": true |
| }, |
| "tiny_attention_adapter": { |
| "gqa_axis_adapter": "transpose query dimensions 2 and 3", |
| "attention_backend": "gpt_oss.triton.attention.attention_ref", |
| "reason": "tiny config uses 1 KV head, 4 groups, and head_dim 32", |
| "official_source_modified": false |
| }, |
| "files": { |
| "full_logits.safetensors": "5dc16765dd175ff95e7a923e1569dcca83880b044bb4de53cadb419c4535018f", |
| "cached_logits.safetensors": "5dc16765dd175ff95e7a923e1569dcca83880b044bb4de53cadb419c4535018f", |
| "routing.safetensors": "f94ffb12b3da2c922fc7e95134603e155fd06046dbaba9610360e8aea2d58311", |
| "expert_projection.safetensors": "33f73333341f7a86b6a44d273581a096bed789d927a90ce7411d43c0437748bb", |
| "greedy_continuation.json": "1478ed0112bf32f0444a5dd1bae9eeaebe2cc751396d1fe92513e6e208984b2d" |
| } |
| } |
|
|
