TeleAI-AI-Flow/AI-FLow-OpticsReview-9B

1{2  "created_utc": "2026-09-23T06:26:19.866579+00:00",3  "source_dtype": "float32",4  "export_dtype": "bfloat16",5  "conversion": "verl DTensor shard merge; explicit BF16 rounding, not FP8/INT quantization",6  "world_size": 6,7  "tensor_count": 760,8  "parameter_count": 9409813744,9  "visual_tensor_count": 333,10  "visual_parameter_count": 456010480,11  "weight_bytes": 18819627488,12  "validation": {13    "architecture_keys_and_shapes": "pass",14    "all_weights_finite": true,15    "serialized_tensors_equal_bf16_merged_source": 760,16    "processor_and_thinking_chat_template": "pass"17  },18  "config_changes": [19    "dtype float32 -> bfloat16",20    "removed hardware-specific FlashAttention3 setting",21    "recorded export transformers version"22  ],23  "torch_version": "2.11.0+cu130",24  "transformers_version": "5.12.1",25  "model": "OpticsReviewer-9B",26  "checkpoint": "global_step_350",27  "license": "cc-by-nc-4.0",28  "evaluation_source": "Supplied manuscript: A learnable paradigm for scientific evaluation: expert judgment encoded in an optics peer-review model. Benchmark values were not remeasured during export.",29  "paper_source_sha256": "0648c4df642011e6ea71fb95b97978befc659f05971b52b4b796d6e3f600870a",30  "source_weight_shards": {31    "model_world_size_6_rank_0.pt": {32      "sha256": "331f7a12d89074824f6aa3e6299b69c13f34253907f2d5653463aba7c588e7de",33      "bytes": 627550746334    },35    "model_world_size_6_rank_1.pt": {36      "sha256": "3ca2d417cb7db46e1396cd4e2b5998242e7449a42255df53fa4f4bdc95dd93a6",37      "bytes": 627550746338    },39    "model_world_size_6_rank_2.pt": {40      "sha256": "b466a5c39c15ec7e4e82a0d6cc72a50a1b40b5f20116ee5175b365b1e27f2cd9",41      "bytes": 627550746342    },43    "model_world_size_6_rank_3.pt": {44      "sha256": "4d64f6d7e4383274cdc1d141ad2da5d041827406fe16dcb6c6a6c5a6d7fc3d2b",45      "bytes": 627550746346    },47    "model_world_size_6_rank_4.pt": {48      "sha256": "fcb1a1fba61999db731a664b3d113cc06a879e50252662c9c3806db136b59a47",49      "bytes": 627550746350    },51    "model_world_size_6_rank_5.pt": {52      "sha256": "66de492df301c7850df3703b8e8b16c7c881313fc57f1708891c23911246b589",53      "bytes": 627550746354    }55  },56  "loading_smoke": {57    "test": "synthetic image-conditioned load/generation smoke; not a benchmark",58    "enable_thinking": true,59    "dtype": "torch.bfloat16",60    "attn_implementation": "sdpa",61    "loading_info": {62      "missing_keys": [],63      "unexpected_keys": [],64      "mismatched_keys": [],65      "error_msgs": []66    },67    "input_tokens": 99,68    "generated_tokens": 64,69    "all_generation_scores_finite": true,70    "max_new_tokens": 64,71    "output_may_be_truncated": true,72    "peak_cuda_allocated_gib": 17.65506792068481473  }74}75