{
  "gptq": [
    {
      "process": "gptq",
      "layer": 0,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000465595",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.838",
      "fwd_time": "1.240",
      "(v)ram": "cuda 2.82G"
    },
    {
      "process": "gptq",
      "layer": 0,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000305803",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.571",
      "fwd_time": "0.517",
      "(v)ram": "cuda 2.96G"
    },
    {
      "process": "gptq",
      "layer": 0,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000691",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.946",
      "fwd_time": "1.219",
      "(v)ram": "cuda 3.12G"
    },
    {
      "process": "gptq",
      "layer": 0,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000002992",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.341",
      "fwd_time": "1.683",
      "(v)ram": "cuda 5.49G"
    },
    {
      "process": "gptq",
      "layer": 0,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000003201",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.540",
      "fwd_time": "1.683",
      "(v)ram": "cuda 5.49G"
    },
    {
      "process": "gptq",
      "layer": 0,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000011",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.401",
      "fwd_time": "3.027",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 1,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000048377",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.311",
      "fwd_time": "0.578",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 1,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000029758",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.544",
      "fwd_time": "0.435",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 1,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000068",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.353",
      "fwd_time": "1.070",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 1,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000006274",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.839",
      "fwd_time": "1.533",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 1,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000006530",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.897",
      "fwd_time": "1.533",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 1,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000016",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.298",
      "fwd_time": "2.843",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 2,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000065660",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.461",
      "fwd_time": "0.458",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 2,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000039966",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.271",
      "fwd_time": "0.472",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 2,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000109",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.008",
      "fwd_time": "1.063",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 2,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000011682",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.524",
      "fwd_time": "1.466",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 2,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000012390",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.542",
      "fwd_time": "1.466",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 2,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000053",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.445",
      "fwd_time": "2.985",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 3,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000071029",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.450",
      "fwd_time": "1.097",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 3,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0001030882",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.632",
      "fwd_time": "1.097",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 3,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000068607",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.771",
      "fwd_time": "1.097",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 3,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000149",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.971",
      "fwd_time": "0.918",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 3,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000013900",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.998",
      "fwd_time": "1.245",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 3,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000014284",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.084",
      "fwd_time": "1.245",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 3,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000040",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.235",
      "fwd_time": "2.701",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 4,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000068506",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.197",
      "fwd_time": "0.622",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 4,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000043613",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.714",
      "fwd_time": "0.473",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 4,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000150",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.891",
      "fwd_time": "1.187",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 4,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000019025",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.847",
      "fwd_time": "1.550",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 4,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000018767",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.865",
      "fwd_time": "1.550",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 4,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000052",
      "samples": "46025",
      "damp": "0.01000",
      "time": "14.583",
      "fwd_time": "2.958",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 5,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000096762",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.360",
      "fwd_time": "0.651",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 5,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000063109",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.823",
      "fwd_time": "0.468",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 5,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000166",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.406",
      "fwd_time": "1.192",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 5,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000023280",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.872",
      "fwd_time": "1.548",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 5,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000024457",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.909",
      "fwd_time": "1.548",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 5,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000076",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.991",
      "fwd_time": "2.915",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 6,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000172638",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.553",
      "fwd_time": "0.603",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 6,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000103571",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.410",
      "fwd_time": "0.443",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 6,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000315",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.838",
      "fwd_time": "1.136",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 6,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000034171",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.269",
      "fwd_time": "1.515",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 6,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000036210",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.413",
      "fwd_time": "1.515",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 6,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000134",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.372",
      "fwd_time": "3.043",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 7,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000059113",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.050",
      "fwd_time": "0.973",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 7,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0000787790",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.114",
      "fwd_time": "0.973",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 7,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000055630",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.235",
      "fwd_time": "0.973",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 7,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000369",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.965",
      "fwd_time": "0.922",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 7,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000038006",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.871",
      "fwd_time": "1.194",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 7,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000040251",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.951",
      "fwd_time": "1.194",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 7,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000171",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.071",
      "fwd_time": "2.691",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 8,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000140466",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.524",
      "fwd_time": "0.484",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 8,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000088766",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.591",
      "fwd_time": "0.478",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 8,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000424",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.995",
      "fwd_time": "1.074",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 8,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000042088",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.027",
      "fwd_time": "1.536",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 8,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000044353",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.224",
      "fwd_time": "1.536",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 8,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000186",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.656",
      "fwd_time": "2.862",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 9,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000146743",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.415",
      "fwd_time": "0.619",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 9,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000090112",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.189",
      "fwd_time": "0.445",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 9,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000407",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.258",
      "fwd_time": "1.194",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 9,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000044021",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.325",
      "fwd_time": "1.812",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 9,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000046012",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.398",
      "fwd_time": "1.812",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 9,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000203",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.828",
      "fwd_time": "2.773",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 10,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000148848",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.323",
      "fwd_time": "0.531",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 10,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000089107",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.390",
      "fwd_time": "0.446",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 10,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000605",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.166",
      "fwd_time": "1.078",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 10,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000044844",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.541",
      "fwd_time": "1.536",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 10,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000046451",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.599",
      "fwd_time": "1.536",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 10,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000244",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.810",
      "fwd_time": "2.948",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 11,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000049339",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.118",
      "fwd_time": "0.903",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 11,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0000630472",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.206",
      "fwd_time": "0.903",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 11,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000047324",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.319",
      "fwd_time": "0.903",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 11,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000001008",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.058",
      "fwd_time": "0.868",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 11,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000052183",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.230",
      "fwd_time": "1.216",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 11,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000054071",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.277",
      "fwd_time": "1.216",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 11,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000347",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.730",
      "fwd_time": "2.601",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 12,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000193344",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.267",
      "fwd_time": "0.489",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 12,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000115214",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.999",
      "fwd_time": "0.527",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 12,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000000890",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.844",
      "fwd_time": "1.209",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 12,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000054587",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.800",
      "fwd_time": "1.546",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 12,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000056580",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.822",
      "fwd_time": "1.546",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 12,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000387",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.711",
      "fwd_time": "2.981",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 13,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000205298",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.015",
      "fwd_time": "0.577",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 13,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000116700",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.442",
      "fwd_time": "0.492",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 13,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000001050",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.942",
      "fwd_time": "1.106",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 13,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000060140",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.864",
      "fwd_time": "1.616",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 13,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000062457",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.944",
      "fwd_time": "1.616",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 13,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000448",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.853",
      "fwd_time": "2.906",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 14,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000228385",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.808",
      "fwd_time": "0.496",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 14,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000134632",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.232",
      "fwd_time": "0.537",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 14,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000001706",
      "samples": "46025",
      "damp": "0.01000",
      "time": "6.292",
      "fwd_time": "1.108",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 14,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000068234",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.267",
      "fwd_time": "1.562",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 14,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000065948",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.368",
      "fwd_time": "1.562",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 14,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000568",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.929",
      "fwd_time": "2.897",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 15,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000042516",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.526",
      "fwd_time": "0.846",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 15,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000042227",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.863",
      "fwd_time": "0.846",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 15,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0000498335",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.935",
      "fwd_time": "0.846",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 15,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000001376",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.051",
      "fwd_time": "0.896",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 15,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000080230",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.382",
      "fwd_time": "1.276",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 15,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000082196",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.410",
      "fwd_time": "1.276",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 15,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000000800",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.850",
      "fwd_time": "2.649",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 16,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000244958",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.882",
      "fwd_time": "0.496",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 16,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000139996",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.429",
      "fwd_time": "0.539",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 16,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000001649",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.950",
      "fwd_time": "1.106",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 16,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000090578",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.231",
      "fwd_time": "1.589",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 16,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000093029",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.275",
      "fwd_time": "1.589",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 16,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000001030",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.792",
      "fwd_time": "2.912",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 17,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000282585",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.239",
      "fwd_time": "0.598",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 17,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000146324",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.718",
      "fwd_time": "0.562",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 17,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000002194",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.066",
      "fwd_time": "1.213",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 17,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000110414",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.310",
      "fwd_time": "1.666",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 17,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000107577",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.355",
      "fwd_time": "1.666",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 17,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000001347",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.906",
      "fwd_time": "3.058",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 18,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000304728",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.571",
      "fwd_time": "0.585",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 18,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000170016",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.615",
      "fwd_time": "0.548",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 18,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000003511",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.093",
      "fwd_time": "1.105",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 18,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000138877",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.130",
      "fwd_time": "1.526",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 18,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000143212",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.158",
      "fwd_time": "1.526",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 18,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000002540",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.289",
      "fwd_time": "2.943",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 19,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000044707",
      "samples": "46025",
      "damp": "0.01000",
      "time": "19.904",
      "fwd_time": "0.856",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 19,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0000538362",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.042",
      "fwd_time": "0.856",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 19,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000048465",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.141",
      "fwd_time": "0.856",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 19,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000008460",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.935",
      "fwd_time": "0.872",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 19,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000218919",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.212",
      "fwd_time": "1.298",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 19,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000211867",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.329",
      "fwd_time": "1.298",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 19,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000004457",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.790",
      "fwd_time": "2.703",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 20,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000626688",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.637",
      "fwd_time": "0.572",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 20,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000359357",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.110",
      "fwd_time": "0.499",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 20,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000007770",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.139",
      "fwd_time": "1.145",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 20,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000283249",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.657",
      "fwd_time": "1.577",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 20,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000268738",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.850",
      "fwd_time": "1.577",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 20,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000006253",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.270",
      "fwd_time": "2.899",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 21,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000755664",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.286",
      "fwd_time": "0.470",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 21,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000491567",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.860",
      "fwd_time": "0.520",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 21,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000008459",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.031",
      "fwd_time": "1.187",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 21,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000305040",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.990",
      "fwd_time": "1.587",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 21,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000320680",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.029",
      "fwd_time": "1.587",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 21,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000005691",
      "samples": "46025",
      "damp": "0.01000",
      "time": "15.314",
      "fwd_time": "2.919",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 22,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000670322",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.261",
      "fwd_time": "0.530",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 22,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000417217",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.178",
      "fwd_time": "0.507",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 22,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000006635",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.132",
      "fwd_time": "1.099",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 22,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000388471",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.702",
      "fwd_time": "1.536",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 22,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000351618",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.860",
      "fwd_time": "1.536",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 22,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000006516",
      "samples": "46025",
      "damp": "0.01000",
      "time": "15.353",
      "fwd_time": "2.998",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 23,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000069196",
      "samples": "46025",
      "damp": "0.01000",
      "time": "19.722",
      "fwd_time": "1.023",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 23,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000077916",
      "samples": "46025",
      "damp": "0.01000",
      "time": "19.821",
      "fwd_time": "1.023",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 23,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0000793075",
      "samples": "46025",
      "damp": "0.01000",
      "time": "19.915",
      "fwd_time": "1.023",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 23,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000011365",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.853",
      "fwd_time": "0.853",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 23,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000435874",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.105",
      "fwd_time": "1.357",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 23,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000463842",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.134",
      "fwd_time": "1.357",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 23,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000008425",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.878",
      "fwd_time": "2.651",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 24,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000698582",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.795",
      "fwd_time": "0.580",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 24,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000404147",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.633",
      "fwd_time": "0.551",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 24,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000012276",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.637",
      "fwd_time": "1.124",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 24,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000538200",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.149",
      "fwd_time": "1.658",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 24,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000504490",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.406",
      "fwd_time": "1.658",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 24,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000010879",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.926",
      "fwd_time": "3.118",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 25,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000758185",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.323",
      "fwd_time": "0.525",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 25,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000412498",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.245",
      "fwd_time": "0.505",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 25,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000013347",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.787",
      "fwd_time": "1.111",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 25,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000522937",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.877",
      "fwd_time": "1.676",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 25,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000513961",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.993",
      "fwd_time": "1.676",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 25,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000012097",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.317",
      "fwd_time": "3.060",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 26,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000698884",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.032",
      "fwd_time": "0.489",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 26,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000367971",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.119",
      "fwd_time": "0.500",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 26,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000021574",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.719",
      "fwd_time": "1.132",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 26,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000521411",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.782",
      "fwd_time": "1.691",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 26,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000537349",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.866",
      "fwd_time": "1.691",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 26,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000015733",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.857",
      "fwd_time": "2.987",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 27,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000077329",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.770",
      "fwd_time": "0.956",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 27,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0000883736",
      "samples": "46025",
      "damp": "0.01000",
      "time": "21.046",
      "fwd_time": "0.956",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 27,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000080222",
      "samples": "46025",
      "damp": "0.01000",
      "time": "21.144",
      "fwd_time": "0.956",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 27,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000039121",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.788",
      "fwd_time": "0.893",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 27,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000675765",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.633",
      "fwd_time": "1.238",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 27,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000635203",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.692",
      "fwd_time": "1.238",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 27,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000022758",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.126",
      "fwd_time": "2.684",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 28,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0000878155",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.772",
      "fwd_time": "0.570",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 28,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000424791",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.331",
      "fwd_time": "0.506",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 28,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000027610",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.341",
      "fwd_time": "1.166",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 28,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000662915",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.470",
      "fwd_time": "1.620",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 28,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000719431",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.574",
      "fwd_time": "1.620",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 28,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000024675",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.238",
      "fwd_time": "3.005",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 29,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0001030115",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.945",
      "fwd_time": "0.585",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 29,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000482778",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.953",
      "fwd_time": "0.520",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 29,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000029918",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.085",
      "fwd_time": "1.145",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 29,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000713329",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.608",
      "fwd_time": "1.606",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 29,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000777051",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.647",
      "fwd_time": "1.606",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 29,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000029918",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.246",
      "fwd_time": "3.027",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 30,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0001099097",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.325",
      "fwd_time": "0.556",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 30,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000528073",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.685",
      "fwd_time": "0.550",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 30,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000046937",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.467",
      "fwd_time": "1.153",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 30,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000766497",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.782",
      "fwd_time": "1.559",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 30,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000854201",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.896",
      "fwd_time": "1.559",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 30,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000040538",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.052",
      "fwd_time": "3.100",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 31,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000117875",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.046",
      "fwd_time": "0.860",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 31,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000146597",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.150",
      "fwd_time": "0.860",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 31,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0001238897",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.185",
      "fwd_time": "0.860",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 31,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000079213",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.054",
      "fwd_time": "0.969",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 31,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000901246",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.966",
      "fwd_time": "1.305",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 31,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001026148",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.995",
      "fwd_time": "1.305",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 31,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000054125",
      "samples": "46025",
      "damp": "0.01000",
      "time": "15.526",
      "fwd_time": "2.702",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 32,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0001411392",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.297",
      "fwd_time": "0.645",
      "(v)ram": "cuda 9.99G"
    },
    {
      "process": "gptq",
      "layer": 32,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000633579",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.290",
      "fwd_time": "0.506",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 32,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000059818",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.378",
      "fwd_time": "1.083",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 32,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001081419",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.008",
      "fwd_time": "1.569",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 32,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001243875",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.039",
      "fwd_time": "1.569",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 32,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000065014",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.164",
      "fwd_time": "2.978",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 33,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0001823122",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.205",
      "fwd_time": "0.526",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 33,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000766750",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.380",
      "fwd_time": "0.463",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 33,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000079385",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.661",
      "fwd_time": "1.103",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 33,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001244953",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.917",
      "fwd_time": "1.548",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 33,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001456266",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.994",
      "fwd_time": "1.548",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 33,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000071197",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.185",
      "fwd_time": "3.071",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 34,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0001931272",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.665",
      "fwd_time": "0.593",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 34,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000884646",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.202",
      "fwd_time": "0.516",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 34,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000076319",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.294",
      "fwd_time": "1.218",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 34,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001349536",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.966",
      "fwd_time": "1.678",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 34,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001622027",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.144",
      "fwd_time": "1.678",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 34,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000092788",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.727",
      "fwd_time": "3.016",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 35,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000297213",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.105",
      "fwd_time": "0.994",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 35,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000202671",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.288",
      "fwd_time": "0.994",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 35,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0002131211",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.312",
      "fwd_time": "0.994",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 35,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000165856",
      "samples": "46025",
      "damp": "0.01000",
      "time": "6.004",
      "fwd_time": "0.920",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 35,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001520905",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.776",
      "fwd_time": "1.281",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 35,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001754468",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.835",
      "fwd_time": "1.281",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 35,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000088191",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.986",
      "fwd_time": "2.571",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 36,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002695622",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.256",
      "fwd_time": "0.575",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 36,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001382797",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.288",
      "fwd_time": "0.509",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 36,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000042044",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.340",
      "fwd_time": "1.161",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 36,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001673946",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.956",
      "fwd_time": "1.592",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 36,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001709744",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.124",
      "fwd_time": "1.592",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 36,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000069419",
      "samples": "46025",
      "damp": "0.01000",
      "time": "19.004",
      "fwd_time": "3.008",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 37,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002804669",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.277",
      "fwd_time": "0.619",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 37,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001635379",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.277",
      "fwd_time": "0.509",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 37,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000043134",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.981",
      "fwd_time": "1.210",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 37,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001598896",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.962",
      "fwd_time": "1.641",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 37,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001610927",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.038",
      "fwd_time": "1.641",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 37,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000061502",
      "samples": "46025",
      "damp": "0.01000",
      "time": "18.387",
      "fwd_time": "3.030",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 38,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002559712",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.407",
      "fwd_time": "0.469",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 38,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001519346",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.260",
      "fwd_time": "0.479",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 38,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000033474",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.072",
      "fwd_time": "1.151",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 38,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002102250",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.566",
      "fwd_time": "1.562",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 38,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001819107",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.609",
      "fwd_time": "1.562",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 38,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000064125",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.788",
      "fwd_time": "2.976",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 39,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000256012",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.503",
      "fwd_time": "1.022",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 39,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0002164532",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.587",
      "fwd_time": "1.022",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 39,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000195338",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.718",
      "fwd_time": "1.022",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 39,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000100446",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.098",
      "fwd_time": "0.867",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 39,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001702656",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.953",
      "fwd_time": "1.309",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 39,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001858761",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.025",
      "fwd_time": "1.309",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 39,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000053651",
      "samples": "46025",
      "damp": "0.01000",
      "time": "15.871",
      "fwd_time": "2.499",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 40,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002165669",
      "samples": "46025",
      "damp": "0.01000",
      "time": "6.048",
      "fwd_time": "0.716",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 40,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001219398",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.178",
      "fwd_time": "0.578",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 40,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000043043",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.662",
      "fwd_time": "1.255",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 40,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002016520",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.306",
      "fwd_time": "1.668",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 40,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001809801",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.418",
      "fwd_time": "1.668",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 40,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000061339",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.490",
      "fwd_time": "2.959",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 41,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002043740",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.530",
      "fwd_time": "0.484",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 41,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001144006",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.321",
      "fwd_time": "0.535",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 41,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000034344",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.390",
      "fwd_time": "1.197",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 41,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001675813",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.233",
      "fwd_time": "1.669",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 41,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001684914",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.310",
      "fwd_time": "1.669",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 41,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000066101",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.020",
      "fwd_time": "3.095",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 42,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0001864982",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.811",
      "fwd_time": "0.757",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 42,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001013298",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.442",
      "fwd_time": "0.610",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 42,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000060871",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.023",
      "fwd_time": "1.175",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 42,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001540212",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.552",
      "fwd_time": "1.625",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 42,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001637856",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.681",
      "fwd_time": "1.625",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 42,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000082843",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.642",
      "fwd_time": "3.006",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 43,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000194965",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.762",
      "fwd_time": "1.057",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 43,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000267600",
      "samples": "46025",
      "damp": "0.01000",
      "time": "20.900",
      "fwd_time": "1.057",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 43,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0002105088",
      "samples": "46025",
      "damp": "0.01000",
      "time": "21.024",
      "fwd_time": "1.057",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 43,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000190630",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.308",
      "fwd_time": "0.961",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 43,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001904914",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.082",
      "fwd_time": "1.307",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 43,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001721011",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.195",
      "fwd_time": "1.307",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 43,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000101658",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.366",
      "fwd_time": "2.625",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 44,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002125827",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.503",
      "fwd_time": "0.515",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 44,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001049343",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.503",
      "fwd_time": "0.529",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 44,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000073438",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.065",
      "fwd_time": "1.094",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 44,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001767046",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.428",
      "fwd_time": "1.571",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 44,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001980982",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.561",
      "fwd_time": "1.571",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 44,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000107053",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.416",
      "fwd_time": "2.982",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 45,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002301205",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.100",
      "fwd_time": "0.545",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 45,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001133411",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.269",
      "fwd_time": "0.524",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 45,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000067958",
      "samples": "46025",
      "damp": "0.01000",
      "time": "5.488",
      "fwd_time": "1.194",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 45,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001815128",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.960",
      "fwd_time": "1.673",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 45,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002038248",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.999",
      "fwd_time": "1.673",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 45,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000117689",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.039",
      "fwd_time": "3.062",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 46,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002410166",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.602",
      "fwd_time": "0.638",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 46,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001195199",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.756",
      "fwd_time": "0.476",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 46,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000106593",
      "samples": "46025",
      "damp": "0.01000",
      "time": "6.061",
      "fwd_time": "1.125",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 46,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001838977",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.260",
      "fwd_time": "1.602",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 46,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002108314",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.330",
      "fwd_time": "1.602",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 46,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000135484",
      "samples": "46025",
      "damp": "0.01000",
      "time": "19.642",
      "fwd_time": "2.899",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 47,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000272281",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.301",
      "fwd_time": "0.954",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 47,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000459376",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.311",
      "fwd_time": "0.954",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 47,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0002806311",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.431",
      "fwd_time": "0.954",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 47,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000264252",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.321",
      "fwd_time": "0.802",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 47,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002192224",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.299",
      "fwd_time": "1.285",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 47,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002510076",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.386",
      "fwd_time": "1.285",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 47,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000174404",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.689",
      "fwd_time": "2.641",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 48,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0002837483",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.622",
      "fwd_time": "0.475",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 48,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001363739",
      "samples": "46025",
      "damp": "0.01000",
      "time": "2.871",
      "fwd_time": "0.479",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 48,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000098116",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.461",
      "fwd_time": "1.083",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 48,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002761787",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.404",
      "fwd_time": "1.466",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 48,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002389715",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.490",
      "fwd_time": "1.466",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 48,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000191603",
      "samples": "46025",
      "damp": "0.01000",
      "time": "11.944",
      "fwd_time": "3.015",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 49,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0003382158",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.781",
      "fwd_time": "0.561",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 49,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001466734",
      "samples": "46025",
      "damp": "0.01000",
      "time": "2.962",
      "fwd_time": "0.482",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 49,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000118660",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.948",
      "fwd_time": "1.109",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 49,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002787075",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.931",
      "fwd_time": "1.614",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 49,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002376087",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.958",
      "fwd_time": "1.614",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 49,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000217703",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.872",
      "fwd_time": "2.958",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 50,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0003170064",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.242",
      "fwd_time": "0.463",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 50,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001498102",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.053",
      "fwd_time": "0.429",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 50,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000222148",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.818",
      "fwd_time": "1.112",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 50,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0002705110",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.625",
      "fwd_time": "1.539",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 50,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0003089729",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.685",
      "fwd_time": "1.539",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 50,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000379229",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.234",
      "fwd_time": "3.088",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 51,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0003904938",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.188",
      "fwd_time": "0.916",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 51,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000367747",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.288",
      "fwd_time": "0.916",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 51,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000770984",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.317",
      "fwd_time": "0.916",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 51,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000675822",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.936",
      "fwd_time": "0.963",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 51,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0003886414",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.197",
      "fwd_time": "1.245",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 51,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0003319729",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.223",
      "fwd_time": "1.245",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 51,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000452783",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.002",
      "fwd_time": "2.651",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 52,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0004915262",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.585",
      "fwd_time": "0.501",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 52,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002357129",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.747",
      "fwd_time": "0.449",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 52,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000185822",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.429",
      "fwd_time": "1.103",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 52,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0003765633",
      "samples": "46025",
      "damp": "0.01000",
      "time": "7.676",
      "fwd_time": "1.590",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 52,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0003840524",
      "samples": "46025",
      "damp": "0.01000",
      "time": "7.829",
      "fwd_time": "1.590",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 52,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000358539",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.909",
      "fwd_time": "3.006",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 53,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0004868115",
      "samples": "46025",
      "damp": "0.01000",
      "time": "2.499",
      "fwd_time": "0.501",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 53,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002424216",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.313",
      "fwd_time": "0.437",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 53,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000161716",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.860",
      "fwd_time": "1.183",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 53,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004000879",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.965",
      "fwd_time": "1.617",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 53,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0003850227",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.024",
      "fwd_time": "1.617",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 53,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000381373",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.209",
      "fwd_time": "2.928",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 54,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0004595161",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.252",
      "fwd_time": "0.461",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 54,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002529994",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.525",
      "fwd_time": "0.448",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 54,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000212727",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.330",
      "fwd_time": "1.125",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 54,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004466511",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.232",
      "fwd_time": "1.593",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 54,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004936524",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.393",
      "fwd_time": "1.593",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 54,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000530516",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.039",
      "fwd_time": "3.019",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 55,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0004352777",
      "samples": "46025",
      "damp": "0.01000",
      "time": "15.776",
      "fwd_time": "0.849",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 55,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000914873",
      "samples": "46025",
      "damp": "0.01000",
      "time": "15.829",
      "fwd_time": "0.849",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 55,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000395639",
      "samples": "46025",
      "damp": "0.01000",
      "time": "15.901",
      "fwd_time": "0.849",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 55,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000424179",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.714",
      "fwd_time": "0.839",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 55,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004478328",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.086",
      "fwd_time": "1.282",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 55,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004768694",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.150",
      "fwd_time": "1.282",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 55,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000382846",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.185",
      "fwd_time": "2.551",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 56,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0004960684",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.595",
      "fwd_time": "0.633",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 56,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002529142",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.049",
      "fwd_time": "0.428",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 56,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000178074",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.998",
      "fwd_time": "1.039",
      "(v)ram": "cuda 10.G"
    },
    {
      "process": "gptq",
      "layer": 56,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004939634",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.351",
      "fwd_time": "1.561",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 56,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004486849",
      "samples": "46025",
      "damp": "0.01000",
      "time": "10.394",
      "fwd_time": "1.561",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 56,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000280172",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.913",
      "fwd_time": "3.123",
      "(v)ram": "cuda 10.01G"
    },
    {
      "process": "gptq",
      "layer": 57,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0005098134",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.675",
      "fwd_time": "0.549",
      "(v)ram": "cuda 10.02G"
    },
    {
      "process": "gptq",
      "layer": 57,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002639756",
      "samples": "46025",
      "damp": "0.01000",
      "time": "6.484",
      "fwd_time": "0.463",
      "(v)ram": "cuda 10.27G"
    },
    {
      "process": "gptq",
      "layer": 57,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000143255",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.131",
      "fwd_time": "1.204",
      "(v)ram": "cuda 10.22G"
    },
    {
      "process": "gptq",
      "layer": 57,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0005055156",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.871",
      "fwd_time": "1.516",
      "(v)ram": "cuda 10.17G"
    },
    {
      "process": "gptq",
      "layer": 57,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0005675868",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.919",
      "fwd_time": "1.516",
      "(v)ram": "cuda 10.17G"
    },
    {
      "process": "gptq",
      "layer": 57,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000293966",
      "samples": "46025",
      "damp": "0.01000",
      "time": "13.795",
      "fwd_time": "3.117",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 58,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0004955187",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.365",
      "fwd_time": "0.545",
      "(v)ram": "cuda 10.08G"
    },
    {
      "process": "gptq",
      "layer": 58,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002751105",
      "samples": "46025",
      "damp": "0.01000",
      "time": "2.776",
      "fwd_time": "0.475",
      "(v)ram": "cuda 10.08G"
    },
    {
      "process": "gptq",
      "layer": 58,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000218265",
      "samples": "46025",
      "damp": "0.01000",
      "time": "2.988",
      "fwd_time": "1.120",
      "(v)ram": "cuda 10.08G"
    },
    {
      "process": "gptq",
      "layer": 58,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006015578",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.663",
      "fwd_time": "1.510",
      "(v)ram": "cuda 10.07G"
    },
    {
      "process": "gptq",
      "layer": 58,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006751317",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.741",
      "fwd_time": "1.510",
      "(v)ram": "cuda 10.07G"
    },
    {
      "process": "gptq",
      "layer": 58,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000488076",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.374",
      "fwd_time": "2.919",
      "(v)ram": "cuda 10.08G"
    },
    {
      "process": "gptq",
      "layer": 59,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000493040",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.238",
      "fwd_time": "0.880",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 59,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0005161758",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.711",
      "fwd_time": "0.880",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 59,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0001953404",
      "samples": "46025",
      "damp": "0.01000",
      "time": "17.753",
      "fwd_time": "0.880",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 59,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0001022928",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.176",
      "fwd_time": "0.769",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 59,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006658686",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.717",
      "fwd_time": "1.271",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 59,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006093653",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.902",
      "fwd_time": "1.271",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 59,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000679011",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.651",
      "fwd_time": "2.495",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 60,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0005593422",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.221",
      "fwd_time": "0.469",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 60,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002729036",
      "samples": "46025",
      "damp": "0.01000",
      "time": "2.679",
      "fwd_time": "0.440",
      "(v)ram": "cuda 10.08G"
    },
    {
      "process": "gptq",
      "layer": 60,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000513208",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.113",
      "fwd_time": "1.059",
      "(v)ram": "cuda 10.08G"
    },
    {
      "process": "gptq",
      "layer": 60,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0005944245",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.187",
      "fwd_time": "1.562",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 60,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006348773",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.241",
      "fwd_time": "1.562",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 60,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0000745988",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.303",
      "fwd_time": "2.963",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 61,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0004403089",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.552",
      "fwd_time": "0.500",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 61,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002546274",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.195",
      "fwd_time": "0.462",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 61,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0000778997",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.322",
      "fwd_time": "1.180",
      "(v)ram": "cuda 10.1G"
    },
    {
      "process": "gptq",
      "layer": 61,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006772048",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.103",
      "fwd_time": "1.634",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 61,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006411800",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.131",
      "fwd_time": "1.634",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 61,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001109535",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.444",
      "fwd_time": "2.945",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 62,
      "module": "linear_attn.in_proj_qkv",
      "feat: in, out": "5120, 10240",
      "dtype: size": "bf16: 112.5MB",
      "loss": "0.0004666405",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.666",
      "fwd_time": "0.466",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 62,
      "module": "linear_attn.in_proj_z",
      "feat: in, out": "5120, 6144",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002469899",
      "samples": "46025",
      "damp": "0.01000",
      "time": "2.752",
      "fwd_time": "0.470",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 62,
      "module": "linear_attn.out_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002129779",
      "samples": "46025",
      "damp": "0.01000",
      "time": "3.847",
      "fwd_time": "1.142",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 62,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006328930",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.672",
      "fwd_time": "1.594",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 62,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0006014320",
      "samples": "46025",
      "damp": "0.01000",
      "time": "8.837",
      "fwd_time": "1.594",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 62,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0001837080",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.573",
      "fwd_time": "2.931",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 63,
      "module": "self_attn.q_proj",
      "feat: in, out": "5120, 12288",
      "dtype: size": "bf16: 135.0MB",
      "loss": "0.0005039408",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.709",
      "fwd_time": "0.928",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 63,
      "module": "self_attn.k_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0000442387",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.856",
      "fwd_time": "0.928",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 63,
      "module": "self_attn.v_proj",
      "feat: in, out": "5120, 1024",
      "dtype: size": "bf16: 11.3MB",
      "loss": "0.0001613512",
      "samples": "46025",
      "damp": "0.01000",
      "time": "16.868",
      "fwd_time": "0.928",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 63,
      "module": "self_attn.o_proj",
      "feat: in, out": "6144, 5120",
      "dtype: size": "bf16: 67.5MB",
      "loss": "0.0002461547",
      "samples": "46025",
      "damp": "0.01000",
      "time": "4.187",
      "fwd_time": "0.948",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 63,
      "module": "mlp.gate_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0005901742",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.777",
      "fwd_time": "1.252",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 63,
      "module": "mlp.up_proj",
      "feat: in, out": "5120, 17408",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0005432940",
      "samples": "46025",
      "damp": "0.01000",
      "time": "9.816",
      "fwd_time": "1.252",
      "(v)ram": "cuda 10.09G"
    },
    {
      "process": "gptq",
      "layer": 63,
      "module": "mlp.down_proj",
      "feat: in, out": "17408, 5120",
      "dtype: size": "bf16: 191.3MB",
      "loss": "0.0004097096",
      "samples": "46025",
      "damp": "0.01000",
      "time": "12.570",
      "fwd_time": "2.643",
      "(v)ram": "cuda 10.09G"
    }
  ]
}