{
  "model_id": "HuggingFaceTB/SmolLM2-135M",
  "model_file": "model_observable_vectors_q4.onnx",
  "model_bytes": 216918299,
  "quantization": {
    "bits": 4,
    "block_size": 32,
    "symmetric": false
  },
  "architecture": {
    "layers": 30,
    "attention_heads": 9,
    "key_value_heads": 3,
    "hidden_size": 576,
    "vocabulary_size": 49152
  },
  "inputs": [
    "input_ids",
    "attention_mask",
    "query_index"
  ],
  "outputs": [
    "next_token_logits",
    "hidden_state_00",
    "hidden_state_01",
    "hidden_state_02",
    "hidden_state_03",
    "hidden_state_04",
    "hidden_state_05",
    "hidden_state_06",
    "hidden_state_07",
    "hidden_state_08",
    "hidden_state_09",
    "hidden_state_10",
    "hidden_state_11",
    "hidden_state_12",
    "hidden_state_13",
    "hidden_state_14",
    "hidden_state_15",
    "hidden_state_16",
    "hidden_state_17",
    "hidden_state_18",
    "hidden_state_19",
    "hidden_state_20",
    "hidden_state_21",
    "hidden_state_22",
    "hidden_state_23",
    "hidden_state_24",
    "hidden_state_25",
    "hidden_state_26",
    "hidden_state_27",
    "hidden_state_28",
    "hidden_state_29",
    "hidden_state_30",
    "attention_01",
    "attention_02",
    "attention_03",
    "attention_04",
    "attention_05",
    "attention_06",
    "attention_07",
    "attention_08",
    "attention_09",
    "attention_10",
    "attention_11",
    "attention_12",
    "attention_13",
    "attention_14",
    "attention_15",
    "attention_16",
    "attention_17",
    "attention_18",
    "attention_19",
    "attention_20",
    "attention_21",
    "attention_22",
    "attention_23",
    "attention_24",
    "attention_25",
    "attention_26",
    "attention_27",
    "attention_28",
    "attention_29",
    "attention_30",
    "attention_contribution_01",
    "attention_contribution_02",
    "attention_contribution_03",
    "attention_contribution_04",
    "attention_contribution_05",
    "attention_contribution_06",
    "attention_contribution_07",
    "attention_contribution_08",
    "attention_contribution_09",
    "attention_contribution_10",
    "attention_contribution_11",
    "attention_contribution_12",
    "attention_contribution_13",
    "attention_contribution_14",
    "attention_contribution_15",
    "attention_contribution_16",
    "attention_contribution_17",
    "attention_contribution_18",
    "attention_contribution_19",
    "attention_contribution_20",
    "attention_contribution_21",
    "attention_contribution_22",
    "attention_contribution_23",
    "attention_contribution_24",
    "attention_contribution_25",
    "attention_contribution_26",
    "attention_contribution_27",
    "attention_contribution_28",
    "attention_contribution_29",
    "attention_contribution_30",
    "counterfactual_hidden_state_01",
    "counterfactual_hidden_state_02",
    "counterfactual_hidden_state_03",
    "counterfactual_hidden_state_04",
    "counterfactual_hidden_state_05",
    "counterfactual_hidden_state_06",
    "counterfactual_hidden_state_07",
    "counterfactual_hidden_state_08",
    "counterfactual_hidden_state_09",
    "counterfactual_hidden_state_10",
    "counterfactual_hidden_state_11",
    "counterfactual_hidden_state_12",
    "counterfactual_hidden_state_13",
    "counterfactual_hidden_state_14",
    "counterfactual_hidden_state_15",
    "counterfactual_hidden_state_16",
    "counterfactual_hidden_state_17",
    "counterfactual_hidden_state_18",
    "counterfactual_hidden_state_19",
    "counterfactual_hidden_state_20",
    "counterfactual_hidden_state_21",
    "counterfactual_hidden_state_22",
    "counterfactual_hidden_state_23",
    "counterfactual_hidden_state_24",
    "counterfactual_hidden_state_25",
    "counterfactual_hidden_state_26",
    "counterfactual_hidden_state_27",
    "counterfactual_hidden_state_28",
    "counterfactual_hidden_state_29",
    "counterfactual_hidden_state_30",
    "attention_output_01",
    "attention_output_02",
    "attention_output_03",
    "attention_output_04",
    "attention_output_05",
    "attention_output_06",
    "attention_output_07",
    "attention_output_08",
    "attention_output_09",
    "attention_output_10",
    "attention_output_11",
    "attention_output_12",
    "attention_output_13",
    "attention_output_14",
    "attention_output_15",
    "attention_output_16",
    "attention_output_17",
    "attention_output_18",
    "attention_output_19",
    "attention_output_20",
    "attention_output_21",
    "attention_output_22",
    "attention_output_23",
    "attention_output_24",
    "attention_output_25",
    "attention_output_26",
    "attention_output_27",
    "attention_output_28",
    "attention_output_29",
    "attention_output_30",
    "mlp_output_01",
    "mlp_output_02",
    "mlp_output_03",
    "mlp_output_04",
    "mlp_output_05",
    "mlp_output_06",
    "mlp_output_07",
    "mlp_output_08",
    "mlp_output_09",
    "mlp_output_10",
    "mlp_output_11",
    "mlp_output_12",
    "mlp_output_13",
    "mlp_output_14",
    "mlp_output_15",
    "mlp_output_16",
    "mlp_output_17",
    "mlp_output_18",
    "mlp_output_19",
    "mlp_output_20",
    "mlp_output_21",
    "mlp_output_22",
    "mlp_output_23",
    "mlp_output_24",
    "mlp_output_25",
    "mlp_output_26",
    "mlp_output_27",
    "mlp_output_28",
    "mlp_output_29",
    "mlp_output_30"
  ],
  "validation_prompt": "Context changes the next token",
  "validation": {
    "outputs": 182,
    "top_10_overlap": 9,
    "manual_vs_transformers_max_logit_error": 3.0517578125e-05,
    "onnx_max_logit_error": 4.329092025756836,
    "last_hidden_mean_cosine": 0.8809197545051575,
    "float_attention_sum_max_error": 6.103515625e-05,
    "float_layer_identity_max_error": 0.0,
    "onnx_attention_sum_max_error": 1.9073486328125e-05,
    "onnx_layer_identity_max_error": 0.0,
    "onnx_counterfactual_max_error": 44.02360534667969,
    "onnx_counterfactual_mean_cosine": 0.9895740747451782,
    "logits_shape": [
      1,
      49152
    ],
    "hidden_shape": [
      1,
      5,
      576
    ],
    "attention_shape": [
      1,
      9,
      5
    ],
    "contribution_shape": [
      1,
      5,
      576
    ],
    "counterfactual_shape": [
      1,
      5,
      576
    ],
    "vector_shape": [
      1,
      576
    ]
  }
}