quantization_config.json

{
  "bits": 4,
  "data_type": "int",
  "group_size": 128,
  "sym": true,
  "iters": 0,
  "autoround_version": "0.12.0",
  "quant_method": "auto-round",
  "packing_format": "auto_round:auto_gptq",
  "extra_config": {
    "model.layers.0.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.1.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.2.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.3.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.4.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.5.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.6.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.7.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.8.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.9.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.10.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.11.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.12.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.13.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.14.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.15.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.16.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.17.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.18.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.19.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.20.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.21.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.22.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.23.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.24.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.25.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.26.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.27.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.28.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.29.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.30.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.31.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.32.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.33.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.34.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.35.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.36.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.37.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.38.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.39.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.40.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.41.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.42.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.43.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.44.self_attn.g_proj": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.3.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.4.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.5.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.6.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.7.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.8.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.9.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.10.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.11.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.12.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.13.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.14.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.15.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.16.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.17.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.18.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.19.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.20.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.21.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.22.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.23.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.24.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.25.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.26.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.27.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.28.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.29.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.30.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.31.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.32.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.33.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.34.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.35.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.36.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.37.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.38.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.39.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.40.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.41.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.42.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.43.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.44.moe.gate": {
      "bits": 16,
      "data_type": "float"
    },
    "model.layers.0.mlp.gate_proj": {
      "bits": 8
    },
    "model.layers.0.mlp.up_proj": {
      "bits": 8
    },
    "model.layers.0.mlp.down_proj": {
      "bits": 8
    },
    "model.layers.1.mlp.gate_proj": {
      "bits": 8
    },
    "model.layers.1.mlp.up_proj": {
      "bits": 8
    },
    "model.layers.1.mlp.down_proj": {
      "bits": 8
    },
    "model.layers.2.mlp.gate_proj": {
      "bits": 8
    },
    "model.layers.2.mlp.up_proj": {
      "bits": 8
    },
    "model.layers.2.mlp.down_proj": {
      "bits": 8
    },
    "model.layers.0.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.0.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.0.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.0.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.1.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.1.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.1.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.1.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.2.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.2.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.2.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.2.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.3.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.3.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.3.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.3.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.4.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.4.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.4.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.4.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.5.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.5.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.5.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.5.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.6.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.6.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.6.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.6.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.7.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.7.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.7.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.7.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.8.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.8.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.8.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.8.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.9.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.9.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.9.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.9.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.10.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.10.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.10.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.10.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.11.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.11.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.11.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.11.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.12.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.12.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.12.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.12.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.13.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.13.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.13.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.13.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.14.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.14.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.14.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.14.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.15.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.15.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.15.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.15.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.16.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.16.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.16.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.16.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.17.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.17.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.17.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.17.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.18.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.18.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.18.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.18.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.19.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.19.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.19.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.19.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.20.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.20.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.20.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.20.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.21.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.21.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.21.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.21.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.22.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.22.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.22.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.22.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.23.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.23.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.23.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.23.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.24.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.24.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.24.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.24.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.25.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.25.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.25.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.25.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.26.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.26.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.26.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.26.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.27.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.27.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.27.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.27.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.28.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.28.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.28.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.28.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.29.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.29.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.29.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.29.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.30.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.30.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.30.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.30.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.31.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.31.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.31.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.31.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.32.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.32.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.32.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.32.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.33.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.33.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.33.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.33.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.34.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.34.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.34.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.34.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.35.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.35.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.35.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.35.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.36.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.36.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.36.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.36.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.37.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.37.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.37.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.37.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.38.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.38.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.38.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.38.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.39.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.39.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.39.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.39.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.40.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.40.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.40.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.40.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.41.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.41.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.41.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.41.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.42.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.42.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.42.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.42.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.43.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.43.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.43.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.43.self_attn.o_proj": {
      "bits": 8
    },
    "model.layers.44.self_attn.q_proj": {
      "bits": 8
    },
    "model.layers.44.self_attn.k_proj": {
      "bits": 8
    },
    "model.layers.44.self_attn.v_proj": {
      "bits": 8
    },
    "model.layers.44.self_attn.o_proj": {
      "bits": 8
    },
    ".*mlp.*": {
      "bits": 8
    },
    ".*self_attn.*": {
      "bits": 8
    },
    ".*layers\\.46.*": {
      "bits": 8
    },
    ".*layers\\.47.*": {
      "bits": 8
    },
    ".*eh_proj.*": {
      "bits": 16,
      "data_type": "float"
    },
    ".*shared_head.*": {
      "bits": 16,
      "data_type": "float"
    },
    ".*layers\\.45.*": {
      "bits": 16,
      "data_type": "float"
    },
    ".*g_proj.*": {
      "bits": 16,
      "data_type": "float"
    },
    ".*moe\\.gate.*": {
      "bits": 16,
      "data_type": "float"
    }
  }
}