{
  "tag": "gpt2g",
  "repo": "onnx-community/gpt2-ONNX-GQA",
  "kind": "text-gqa",
  "notas": "GPT-2 com atencao agrupada (GQA) exportada ONNX/OGA",
  "matrix": "transformer.wte.weight",
  "shape": [
    50257,
    768
  ],
  "dtype": "1",
  "K_experts": 32,
  "K_xi": 8,
  "d_src": 768,
  "n_adapters": 4,
  "inertia": 4229.31640625,
  "samples": 6000,
  "int4": true,
  "adapter_rank": 48,
  "epoca": "<= 2024/2025",
  "gqa_signature": {
    "gqa_nodes": 12,
    "ops_top": [
      [
        "Add",
        97
      ],
      [
        "MatMul",
        73
      ],
      [
        "LayerNormalization",
        25
      ],
      [
        "GroupQueryAttention",
        12
      ],
      [
        "Gelu",
        12
      ],
      [
        "Gather",
        3
      ],
      [
        "Constant",
        2
      ],
      [
        "Cast",
        2
      ]
    ],
    "kv_candidate_shapes": []
  }
}