| { |
| "base_model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", |
| "format": "dashq-packed-linear", |
| "format_version": 1, |
| "method": "dashq", |
| "model_class": "causal_lm", |
| "params": { |
| "bits": 3, |
| "group_size": 32, |
| "low_memory_optimization": false, |
| "moe_hessian_scope": "shared", |
| "n_samples": 128, |
| "scale_zero_dtype": "float16", |
| "symmetric": false, |
| "use_error_compensation": true, |
| "use_optimal_shrinkage": true, |
| "use_weighted_quantization": true |
| }, |
| "quantized_modules": { |
| "model.layers.0.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.0.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.0.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.0.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.0.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.0.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.0.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.1.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.1.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.1.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.1.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.1.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.1.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.1.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.10.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.10.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.10.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.10.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.10.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.10.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.10.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.11.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.11.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.11.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.11.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.11.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.11.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.11.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.12.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.12.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.12.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.12.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.12.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.12.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.12.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.13.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.13.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.13.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.13.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.13.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.13.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.13.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.14.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.14.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.14.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.14.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.14.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.14.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.14.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.15.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.15.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.15.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.15.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.15.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.15.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.15.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.16.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.16.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.16.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.16.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.16.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.16.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.16.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.17.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.17.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.17.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.17.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.17.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.17.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.17.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.18.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.18.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.18.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.18.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.18.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.18.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.18.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.19.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.19.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.19.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.19.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.19.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.19.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.19.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.2.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.2.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.2.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.2.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.2.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.2.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.2.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.20.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.20.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.20.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.20.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.20.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.20.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.20.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.21.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.21.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.21.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.21.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.21.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.21.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.21.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.22.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.22.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.22.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.22.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.22.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.22.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.22.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.23.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.23.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.23.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.23.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.23.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.23.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.23.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.24.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.24.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.24.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.24.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.24.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.24.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.24.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.25.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.25.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.25.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.25.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.25.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.25.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.25.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.26.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.26.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.26.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.26.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.26.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.26.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.26.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.27.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.27.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.27.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.27.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.27.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.27.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.27.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.28.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.28.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.28.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.28.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.28.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.28.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.28.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.29.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.29.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.29.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.29.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.29.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.29.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.29.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.3.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.3.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.3.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.3.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.3.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.3.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.3.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.30.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.30.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.30.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.30.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.30.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.30.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.30.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.31.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.31.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.31.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.31.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.31.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.31.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.31.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.32.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.32.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.32.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.32.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.32.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.32.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.32.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.33.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.33.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.33.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.33.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.33.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.33.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.33.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.34.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.34.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.34.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.34.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.34.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.34.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.34.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.35.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.35.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.35.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.35.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.35.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.35.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.35.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.36.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.36.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.36.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.36.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.36.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.36.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.36.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.37.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.37.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.37.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.37.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.37.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.37.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.37.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.38.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.38.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.38.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.38.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.38.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.38.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.38.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.39.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.39.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.39.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.39.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.39.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.39.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.39.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.4.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.4.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.4.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.4.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.4.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.4.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.4.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.40.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.40.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.40.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.40.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.40.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.40.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.40.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.41.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.41.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.41.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.41.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.41.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.41.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.41.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.42.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.42.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.42.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.42.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.42.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.42.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.42.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.43.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.43.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.43.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.43.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.43.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.43.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.43.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.44.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.44.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.44.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.44.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.44.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.44.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.44.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.45.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.45.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.45.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.45.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.45.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.45.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.45.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.46.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.46.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.46.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.46.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.46.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.46.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.46.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.47.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.47.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.47.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.47.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.47.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.47.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.47.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.48.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.48.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.48.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.48.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.48.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.48.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.48.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.49.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.49.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.49.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.49.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.49.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.49.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.49.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.5.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.5.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.5.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.5.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.5.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.5.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.5.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.50.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.50.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.50.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.50.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.50.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.50.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.50.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.51.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.51.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.51.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.51.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.51.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.51.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.51.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.52.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.52.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.52.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.52.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.52.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.52.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.52.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.53.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.53.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.53.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.53.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.53.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.53.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.53.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.54.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.54.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.54.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.54.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.54.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.54.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.54.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.55.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.55.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.55.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.55.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.55.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.55.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.55.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.56.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.56.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.56.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.56.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.56.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.56.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.56.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.57.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.57.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.57.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.57.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.57.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.57.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.57.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.58.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.58.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.58.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.58.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.58.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.58.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.58.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.59.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.59.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.59.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.59.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.59.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.59.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.59.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.6.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.6.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.6.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.6.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.6.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.6.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.6.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.60.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.60.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.60.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.60.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.60.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.60.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.60.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.61.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.61.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.61.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.61.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.61.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.61.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.61.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.62.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.62.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.62.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.62.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.62.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.62.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.62.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.63.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.63.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.63.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.63.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.63.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.63.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.63.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.7.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.7.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.7.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.7.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.7.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.7.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.7.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.8.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.8.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.8.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.8.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.8.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.8.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.8.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.9.mlp.down_proj": { |
| "group_size": 32, |
| "in_features": 27648, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 27648, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.9.mlp.gate_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.9.mlp.up_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 4423680, |
| "out_features": 27648, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.9.self_attn.k_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.9.self_attn.o_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.9.self_attn.q_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 819200, |
| "out_features": 5120, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| }, |
| "model.layers.9.self_attn.v_proj": { |
| "group_size": 32, |
| "in_features": 5120, |
| "linear_dtype": "bfloat16", |
| "nbits": 3, |
| "num_groups": 163840, |
| "out_features": 1024, |
| "packing": "int3_packed_u32", |
| "quant_in_features": 5120, |
| "runtime_backend": "torch", |
| "scale_zero_dtype": "float16" |
| } |
| }, |
| "results": { |
| "Method": "dashq", |
| "Model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", |
| "ModelSizeGB": 19.499650968, |
| "OriginalSizeGB": 65.527841532, |
| "PPL": 7.0394487380981445, |
| "Params": "{'bits': 3, 'group_size': 32, 'scale_zero_dtype': 'float16', 'n_samples': 128, 'moe_hessian_scope': 'shared', 'use_error_compensation': True, 'use_optimal_shrinkage': True, 'use_weighted_quantization': True, 'symmetric': False, 'low_memory_optimization': False}", |
| "QuantTime": 5324.00167632103, |
| "arc_challenge": 57.508532423208194, |
| "arc_easy": 79.41919191919192, |
| "commonsense_qa": 84.27518427518427, |
| "hellaswag": 80.68113921529576, |
| "lambada_openai": 68.56200271686397, |
| "openbookqa": 46.2, |
| "piqa": 79.27094668117519, |
| "truthfulqa_mc2": 57.96621985458474, |
| "winogrande": 74.9802683504341, |
| "zeroshot_avg": 69.87372060399314 |
| } |
| } |
