{
  "torch_version": "2.14.0+cpu",
  "device": "cpu",
  "upstream_source": {
    "url": "https://raw.githubusercontent.com/huggingface/transformers/d4804f9d76b8c071b993d813e9e87067b780cbee/src/transformers/models/glm5_next/modeling_glm5_next.py",
    "sha256": "78e23c624010751f1578254145f313db2f3266579a0f9a396a3ff6cb5779f43d",
    "local_snapshot": null
  },
  "chunk_comparisons": 64,
  "sequence_lengths": [
    1,
    3,
    4,
    5,
    7,
    8,
    9,
    65
  ],
  "pool_visibility_and_tail_comparisons": 72,
  "weighted_and_mixed_pooling": {
    "nonuniform_hand_example": 1,
    "mixed_batch_pool_slots": 6,
    "mixed_batch_multiquery_tails": 9,
    "includes_all_padding_request": true
  },
  "chunk_sizes": [
    1,
    4,
    8,
    64
  ],
  "normalization_modes": [
    false,
    true
  ],
  "max_chunk_output_absolute_error": 2.980232238769531e-07,
  "max_chunk_final_state_absolute_error": 2.980232238769531e-07,
  "max_scalar_output_absolute_error": 2.0717726612051734e-07,
  "max_continuation_output_absolute_error": 1.7881393432617188e-07,
  "scope": "Six extracted upstream mathematical CPU fallback functions/methods; no model weights, full layer, CUDA/FLA kernel, packed documents, or deployment."
}
