{
  "quantization": {
    "method": "jang-importance",
    "profile": "JANG_4M",
    "target_bits": 4.0,
    "actual_bits": 4.12,
    "block_size": 64,
    "calibration_method": "weights",
    "quantization_method": "mse",
    "scoring_method": "weight-magnitude",
    "bit_widths_used": [
      4,
      8
    ],
    "passthrough_bit_widths_used": [],
    "passthrough_tensor_count": 0,
    "quantization_scheme": "asymmetric",
    "quantization_backend": "mx.quantize",
    "hadamard_rotation": false,
    "mlp_asymmetry_floor": true,
    "split_gate_up_quant": false,
    "split_gate_bits": null,
    "split_up_bits": null,
    "n2_down_bits": null,
    "n2_linear_attn_input_bits": null,
    "n2_linear_attn_out_bits": null,
    "n2_self_attn_bits": null,
    "n2_shared_expert_gate_up_bits": null,
    "n2_token_io_bits": null,
    "n2_lm_head_bits": null,
    "expert_prune_keep": null,
    "expert_prune_map": null,
    "expert_prune_method": null,
    "tensor_quantization_manifest_schema": 1,
    "tensor_quantization_manifest_count": 169,
    "tensor_quantization_manifest": {
      "language_model.model.embed_tokens": {
        "source_tensor": "model.language_model.embed_tokens.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.embed_tokens.weight",
        "scales_key": "language_model.model.embed_tokens.scales",
        "biases_key": "language_model.model.embed_tokens.biases",
        "weight_shape": [
          128000,
          256
        ],
        "scales_shape": [
          128000,
          32
        ],
        "biases_shape": [
          128000,
          32
        ]
      },
      "language_model.model.layers.0.conv.in_proj": {
        "source_tensor": "model.language_model.layers.0.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.0.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.0.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.0.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.0.conv.out_proj": {
        "source_tensor": "model.language_model.layers.0.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.0.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.0.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.0.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.0.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.0.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.0.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.0.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.0.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.0.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.0.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.0.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.0.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.0.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.0.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.0.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.0.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.0.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.0.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.1.conv.in_proj": {
        "source_tensor": "model.language_model.layers.1.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.1.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.1.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.1.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.1.conv.out_proj": {
        "source_tensor": "model.language_model.layers.1.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.1.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.1.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.1.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.1.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.1.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.1.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.1.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.1.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.1.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.1.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.1.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.1.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.1.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.1.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.1.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.1.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.1.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.1.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.10.conv.in_proj": {
        "source_tensor": "model.language_model.layers.10.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.10.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.10.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.10.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.10.conv.out_proj": {
        "source_tensor": "model.language_model.layers.10.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.10.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.10.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.10.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.10.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.10.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.10.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.10.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.10.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.10.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.10.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.10.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.10.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.10.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.10.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.10.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.10.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.10.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.10.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.11.conv.in_proj": {
        "source_tensor": "model.language_model.layers.11.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.11.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.11.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.11.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.11.conv.out_proj": {
        "source_tensor": "model.language_model.layers.11.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.11.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.11.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.11.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.11.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.11.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.11.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.11.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.11.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.11.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.11.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.11.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.11.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.11.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.11.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.11.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.11.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.11.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.11.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.12.conv.in_proj": {
        "source_tensor": "model.language_model.layers.12.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.12.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.12.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.12.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.12.conv.out_proj": {
        "source_tensor": "model.language_model.layers.12.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.12.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.12.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.12.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.12.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.12.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.12.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.12.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.12.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.12.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.12.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.12.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.12.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.12.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.12.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.12.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.12.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.12.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.12.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.13.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.13.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.13.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.13.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.13.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.13.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.13.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.13.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.13.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.13.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.13.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.13.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.13.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.13.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.13.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.13.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.13.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.13.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.13.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.13.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.13.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.13.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.13.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.13.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.13.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.13.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.13.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.13.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.13.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.13.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.13.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.13.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.13.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.13.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.13.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.14.conv.in_proj": {
        "source_tensor": "model.language_model.layers.14.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.14.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.14.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.14.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.14.conv.out_proj": {
        "source_tensor": "model.language_model.layers.14.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.14.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.14.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.14.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.14.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.14.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.14.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.14.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.14.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.14.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.14.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.14.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.14.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.14.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.14.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.14.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.14.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.14.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.14.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.15.conv.in_proj": {
        "source_tensor": "model.language_model.layers.15.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.15.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.15.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.15.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.15.conv.out_proj": {
        "source_tensor": "model.language_model.layers.15.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.15.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.15.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.15.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.15.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.15.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.15.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.15.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.15.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.15.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.15.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.15.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.15.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.15.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.15.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.15.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.15.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.15.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.15.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.16.conv.in_proj": {
        "source_tensor": "model.language_model.layers.16.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.16.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.16.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.16.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.16.conv.out_proj": {
        "source_tensor": "model.language_model.layers.16.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.16.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.16.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.16.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.16.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.16.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.16.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.16.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.16.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.16.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.16.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.16.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.16.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.16.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.16.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.16.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.16.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.16.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.16.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.17.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.17.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.17.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.17.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.17.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.17.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.17.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.17.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.17.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.17.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.17.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.17.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.17.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.17.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.17.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.17.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.17.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.17.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.17.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.17.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.17.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.17.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.17.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.17.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.17.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.17.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.17.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.17.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.17.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.17.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.17.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.17.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.17.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.17.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.17.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.18.conv.in_proj": {
        "source_tensor": "model.language_model.layers.18.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.18.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.18.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.18.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.18.conv.out_proj": {
        "source_tensor": "model.language_model.layers.18.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.18.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.18.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.18.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.18.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.18.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.18.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.18.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.18.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.18.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.18.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.18.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.18.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.18.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.18.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.18.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.18.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.18.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.18.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.19.conv.in_proj": {
        "source_tensor": "model.language_model.layers.19.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.19.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.19.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.19.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.19.conv.out_proj": {
        "source_tensor": "model.language_model.layers.19.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.19.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.19.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.19.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.19.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.19.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.19.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.19.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.19.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.19.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.19.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.19.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.19.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.19.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.19.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.19.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.19.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.19.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.19.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.2.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.2.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.2.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.2.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.2.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.2.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.2.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.2.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.2.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.2.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.2.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.2.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.2.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.2.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.2.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.2.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.2.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.2.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.2.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.2.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.2.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.2.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.2.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.2.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.2.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.2.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.2.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.2.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.2.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.2.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.2.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.2.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.2.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.2.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.2.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.20.conv.in_proj": {
        "source_tensor": "model.language_model.layers.20.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.20.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.20.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.20.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.20.conv.out_proj": {
        "source_tensor": "model.language_model.layers.20.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.20.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.20.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.20.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.20.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.20.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.20.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.20.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.20.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.20.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.20.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.20.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.20.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.20.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.20.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.20.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.20.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.20.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.20.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.21.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.21.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.21.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.21.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.21.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.21.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.21.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.21.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.21.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.21.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.21.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.21.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.21.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.21.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.21.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.21.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.21.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.21.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.21.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.21.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.21.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.21.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.21.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.21.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.21.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.21.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.21.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.21.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.21.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.21.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.21.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.21.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.21.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.21.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.21.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.22.conv.in_proj": {
        "source_tensor": "model.language_model.layers.22.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.22.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.22.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.22.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.22.conv.out_proj": {
        "source_tensor": "model.language_model.layers.22.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.22.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.22.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.22.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.22.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.22.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.22.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.22.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.22.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.22.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.22.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.22.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.22.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.22.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.22.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.22.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.22.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.22.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.22.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.23.conv.in_proj": {
        "source_tensor": "model.language_model.layers.23.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.23.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.23.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.23.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.23.conv.out_proj": {
        "source_tensor": "model.language_model.layers.23.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.23.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.23.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.23.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.23.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.23.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.23.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.23.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.23.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.23.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.23.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.23.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.23.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.23.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.23.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.23.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.23.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.23.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.23.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.24.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.24.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.24.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.24.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.24.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.24.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.24.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.24.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.24.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.24.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.24.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.24.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.24.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.24.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.24.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.24.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.24.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.24.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.24.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.24.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.24.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.24.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.24.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.24.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.24.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.24.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.24.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.24.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.24.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.24.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.24.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.24.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.24.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.24.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.24.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.25.conv.in_proj": {
        "source_tensor": "model.language_model.layers.25.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.25.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.25.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.25.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.25.conv.out_proj": {
        "source_tensor": "model.language_model.layers.25.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.25.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.25.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.25.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.25.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.25.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.25.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.25.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.25.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.25.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.25.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.25.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.25.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.25.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.25.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.25.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.25.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.25.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.25.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.26.conv.in_proj": {
        "source_tensor": "model.language_model.layers.26.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.26.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.26.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.26.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.26.conv.out_proj": {
        "source_tensor": "model.language_model.layers.26.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.26.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.26.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.26.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.26.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.26.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.26.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.26.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.26.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.26.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.26.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.26.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.26.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.26.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.26.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.26.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.26.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.26.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.26.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.27.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.27.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.27.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.27.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.27.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.27.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.27.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.27.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.27.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.27.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.27.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.27.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.27.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.27.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.27.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.27.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.27.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.27.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.27.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.27.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.27.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.27.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.27.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.27.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.27.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.27.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.27.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.27.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.27.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.27.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.27.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.27.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.27.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.27.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.27.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.28.conv.in_proj": {
        "source_tensor": "model.language_model.layers.28.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.28.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.28.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.28.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.28.conv.out_proj": {
        "source_tensor": "model.language_model.layers.28.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.28.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.28.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.28.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.28.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.28.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.28.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.28.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.28.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.28.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.28.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.28.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.28.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.28.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.28.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.28.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.28.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.28.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.28.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.29.conv.in_proj": {
        "source_tensor": "model.language_model.layers.29.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.29.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.29.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.29.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.29.conv.out_proj": {
        "source_tensor": "model.language_model.layers.29.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.29.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.29.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.29.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.29.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.29.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.29.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.29.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.29.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.29.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.29.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.29.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.29.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.29.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.29.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.29.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.29.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.29.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.29.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.3.conv.in_proj": {
        "source_tensor": "model.language_model.layers.3.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.3.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.3.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.3.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.3.conv.out_proj": {
        "source_tensor": "model.language_model.layers.3.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.3.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.3.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.3.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.3.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.3.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.3.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.3.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.3.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.3.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.3.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.3.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.3.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.3.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.3.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.3.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.3.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.3.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.3.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.4.conv.in_proj": {
        "source_tensor": "model.language_model.layers.4.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.4.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.4.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.4.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.4.conv.out_proj": {
        "source_tensor": "model.language_model.layers.4.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.4.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.4.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.4.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.4.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.4.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.4.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.4.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.4.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.4.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.4.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.4.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.4.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.4.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.4.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.4.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.4.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.4.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.4.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.5.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.5.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.5.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.5.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.5.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.5.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.5.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.5.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.5.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.5.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.5.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.5.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.5.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.5.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.5.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.5.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.5.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.5.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.5.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.5.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.5.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.5.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.5.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.5.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.5.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.5.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.5.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.5.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.5.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.5.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.5.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.5.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.5.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.5.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.5.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.6.conv.in_proj": {
        "source_tensor": "model.language_model.layers.6.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.6.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.6.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.6.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.6.conv.out_proj": {
        "source_tensor": "model.language_model.layers.6.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.6.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.6.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.6.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.6.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.6.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.6.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.6.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.6.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.6.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.6.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.6.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.6.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.6.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.6.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.6.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.6.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.6.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.6.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.7.conv.in_proj": {
        "source_tensor": "model.language_model.layers.7.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.7.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.7.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.7.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.7.conv.out_proj": {
        "source_tensor": "model.language_model.layers.7.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.7.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.7.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.7.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.7.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.7.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.7.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.7.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.7.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.7.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.7.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.7.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.7.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.7.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.7.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.7.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.7.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.7.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.7.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.8.conv.in_proj": {
        "source_tensor": "model.language_model.layers.8.conv.in_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.8.conv.in_proj.weight",
        "scales_key": "language_model.model.layers.8.conv.in_proj.scales",
        "biases_key": "language_model.model.layers.8.conv.in_proj.biases",
        "weight_shape": [
          6144,
          256
        ],
        "scales_shape": [
          6144,
          32
        ],
        "biases_shape": [
          6144,
          32
        ]
      },
      "language_model.model.layers.8.conv.out_proj": {
        "source_tensor": "model.language_model.layers.8.conv.out_proj.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.8.conv.out_proj.weight",
        "scales_key": "language_model.model.layers.8.conv.out_proj.scales",
        "biases_key": "language_model.model.layers.8.conv.out_proj.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.8.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.8.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.8.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.8.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.8.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.8.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.8.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.8.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.8.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.8.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.8.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.8.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.8.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.8.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.8.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.9.feed_forward.w1": {
        "source_tensor": "model.language_model.layers.9.feed_forward.w1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.9.feed_forward.w1.weight",
        "scales_key": "language_model.model.layers.9.feed_forward.w1.scales",
        "biases_key": "language_model.model.layers.9.feed_forward.w1.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.9.feed_forward.w2": {
        "source_tensor": "model.language_model.layers.9.feed_forward.w2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.9.feed_forward.w2.weight",
        "scales_key": "language_model.model.layers.9.feed_forward.w2.scales",
        "biases_key": "language_model.model.layers.9.feed_forward.w2.biases",
        "weight_shape": [
          2048,
          1344
        ],
        "scales_shape": [
          2048,
          168
        ],
        "biases_shape": [
          2048,
          168
        ]
      },
      "language_model.model.layers.9.feed_forward.w3": {
        "source_tensor": "model.language_model.layers.9.feed_forward.w3.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "language_model.model.layers.9.feed_forward.w3.weight",
        "scales_key": "language_model.model.layers.9.feed_forward.w3.scales",
        "biases_key": "language_model.model.layers.9.feed_forward.w3.biases",
        "weight_shape": [
          10752,
          256
        ],
        "scales_shape": [
          10752,
          32
        ],
        "biases_shape": [
          10752,
          32
        ]
      },
      "language_model.model.layers.9.self_attn.k_proj": {
        "source_tensor": "model.language_model.layers.9.self_attn.k_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.9.self_attn.k_proj.weight",
        "scales_key": "language_model.model.layers.9.self_attn.k_proj.scales",
        "biases_key": "language_model.model.layers.9.self_attn.k_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "language_model.model.layers.9.self_attn.out_proj": {
        "source_tensor": "model.language_model.layers.9.self_attn.out_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.9.self_attn.out_proj.weight",
        "scales_key": "language_model.model.layers.9.self_attn.out_proj.scales",
        "biases_key": "language_model.model.layers.9.self_attn.out_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.9.self_attn.q_proj": {
        "source_tensor": "model.language_model.layers.9.self_attn.q_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.9.self_attn.q_proj.weight",
        "scales_key": "language_model.model.layers.9.self_attn.q_proj.scales",
        "biases_key": "language_model.model.layers.9.self_attn.q_proj.biases",
        "weight_shape": [
          2048,
          512
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      },
      "language_model.model.layers.9.self_attn.v_proj": {
        "source_tensor": "model.language_model.layers.9.self_attn.v_proj.weight",
        "bits": 8,
        "group_size": 64,
        "weight_key": "language_model.model.layers.9.self_attn.v_proj.weight",
        "scales_key": "language_model.model.layers.9.self_attn.v_proj.scales",
        "biases_key": "language_model.model.layers.9.self_attn.v_proj.biases",
        "weight_shape": [
          512,
          512
        ],
        "scales_shape": [
          512,
          32
        ],
        "biases_shape": [
          512,
          32
        ]
      },
      "model.multi_modal_projector.linear_1": {
        "source_tensor": "model.multi_modal_projector.linear_1.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "model.multi_modal_projector.linear_1.weight",
        "scales_key": "model.multi_modal_projector.linear_1.scales",
        "biases_key": "model.multi_modal_projector.linear_1.biases",
        "weight_shape": [
          2048,
          576
        ],
        "scales_shape": [
          2048,
          72
        ],
        "biases_shape": [
          2048,
          72
        ]
      },
      "model.multi_modal_projector.linear_2": {
        "source_tensor": "model.multi_modal_projector.linear_2.weight",
        "bits": 4,
        "group_size": 64,
        "weight_key": "model.multi_modal_projector.linear_2.weight",
        "scales_key": "model.multi_modal_projector.linear_2.scales",
        "biases_key": "model.multi_modal_projector.linear_2.biases",
        "weight_shape": [
          2048,
          256
        ],
        "scales_shape": [
          2048,
          32
        ],
        "biases_shape": [
          2048,
          32
        ]
      }
    }
  },
  "source_model": {
    "name": "LFM2.5-VL-3B",
    "dtype": "bfloat16",
    "parameters": "2.7B"
  },
  "architecture": {
    "type": "transformer",
    "attention": "gqa",
    "has_vision": true,
    "has_ssm": false,
    "has_moe": false
  },
  "runtime": {
    "total_weight_bytes": 2392606560,
    "total_weight_gb": 2.23
  },
  "capabilities": {
    "reasoning_parser": "qwen3",
    "tool_parser": "lfm2",
    "think_in_template": false,
    "supports_tools": true,
    "supports_thinking": false,
    "family": "lfm2",
    "modality": "vision",
    "modalities": {
      "text": true,
      "vision": true,
      "audio": false,
      "video": false
    },
    "has_vision": true,
    "has_audio": false,
    "has_video": false,
    "cache_type": "hybrid",
    "default_reasoning": "off"
  },
  "format": "jang",
  "format_version": "2.0",
  "chat": {
    "sampling_defaults": {
      "temperature": 0.2,
      "top_p": 1.0,
      "top_k": 50,
      "repetition_penalty": 1.0,
      "source": "vendor_card+generation_config_2026-08-12"
    },
    "stop_token_ids": [
      124900
    ]
  },
  "reasoning": {
    "supported": false,
    "default": "off",
    "think_in_template": false,
    "tiers": null,
    "note": "LFM2.5-VL-3B answers directly; the card states it does not reason. The template has no <think> prefill and no enable_thinking kwarg. `preserve_thinking` (default false) only controls whether <think> in PRIOR assistant turns is kept \u2014 it is history handling, not a reasoning rail."
  },
  "tools": {
    "supported": true,
    "parser": "lfm2",
    "mlx_lm_parser": "pythonic",
    "dialect": "pythonic",
    "format": "<|tool_call_start|>[fn(arg='v')]<|tool_call_end|>",
    "tools_in_system_prompt": true
  },
  "vision": {
    "supported": true,
    "tower": "siglip2_vision_model (so400m-patch16-naflex)",
    "tower_precision": "source (not quantized)",
    "image_token_id": 124907,
    "max_image_tokens": 256,
    "max_tiles": 10,
    "processor": "Lfm2VlProcessor (processor_config.json)"
  }
}