diff --git "a/config.json" "b/config.json" new file mode 100644--- /dev/null +++ "b/config.json" @@ -0,0 +1,5956 @@ +{ + "architectures": [ + "Gemma3ForConditionalGeneration" + ], + "boi_token_index": 255999, + "eoi_token_index": 256000, + "eos_token_id": [ + 1, + 106 + ], + "image_token_index": 262144, + "initializer_range": 0.02, + "mm_tokens_per_image": 256, + "model_type": "gemma3", + "quantization_config": { + "include_input_output_embeddings": false, + "modules_to_not_convert": null, + "quant_method": "torchao", + "quant_type": { + "default": { + "_data": { + "module_fqn_to_config": { + "language_model.layers.0.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.0.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.0.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.0.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.1.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.1.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.1.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.1.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.10.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.10.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.10.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.10.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.11.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.11.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.11.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.11.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.12.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.12.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.12.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.12.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.13.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.13.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.13.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.13.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.14.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.14.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.14.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.14.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.15.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.15.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.15.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.15.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.16.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.16.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.16.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.16.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.17.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.17.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.17.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.17.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.18.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.18.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.18.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.18.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.19.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.19.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.19.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.19.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.2.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.2.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.2.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.2.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.20.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.20.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.20.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.20.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.21.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.21.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.21.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.21.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.22.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.22.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.22.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.22.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.23.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.23.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.23.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.23.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.24.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.24.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.24.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.24.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.25.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.25.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.25.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.25.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.26.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.26.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.26.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.26.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.27.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.27.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.27.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.27.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.28.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.28.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.28.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.28.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.29.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.29.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.29.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.29.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.3.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.3.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.3.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.3.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.30.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.30.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.30.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.30.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.31.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.31.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.31.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.31.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.32.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.32.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.32.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.32.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.33.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.33.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.33.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.33.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.34.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.34.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.34.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.34.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.35.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.35.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.35.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.35.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.36.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.36.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.36.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.36.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.37.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.37.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.37.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.37.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.38.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.38.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.38.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.38.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.39.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.39.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.39.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.39.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.4.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.4.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.4.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.4.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.40.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.40.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.40.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.40.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.41.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.41.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.41.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.41.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.42.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.42.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.42.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.42.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.43.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.43.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.43.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.43.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.44.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.44.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.44.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.44.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.45.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.45.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.45.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.45.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.46.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.46.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.46.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.46.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.5.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.5.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.5.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.5.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.6.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.6.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.6.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.6.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.7.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.7.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.7.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.7.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.8.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.8.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.8.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.8.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.9.mlp.down_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.9.mlp.gate_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.9.mlp.gate_up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + }, + "language_model.layers.9.mlp.up_proj": { + "_data": { + "base_config": { + "_data": { + "group_size": 128, + "int4_choose_qparams_algorithm": "hqq", + "int4_packing_format": "tile_packed_to_4d", + "layout": { + "_data": { + "inner_k_tiles": 8 + }, + "_type": "TensorCoreTiledLayout", + "_version": 1 + }, + "preserve_zero": null, + "set_inductor_config": true, + "use_hqq": false, + "zero_point_domain": { + "_data": "NONE", + "_type": "ZeroPointDomain" + } + }, + "_type": "Int4WeightOnlyConfig", + "_version": 2 + }, + "scale_search_space_size": 20, + "step": "prepare_for_loading" + }, + "_type": "AWQConfig", + "_version": 1 + } + } + }, + "_type": "ModuleFqnToConfig", + "_version": 1 + } + }, + "quant_type_kwargs": {}, + "untie_embedding_weights": false + }, + "text_config": { + "_sliding_window_pattern": 6, + "attention_bias": false, + "attention_dropout": 0.0, + "attn_logit_softcapping": null, + "final_logit_softcapping": null, + "head_dim": 256, + "hidden_activation": "gelu_pytorch_tanh", + "hidden_size": 3840, + "initializer_range": 0.02, + "intermediate_size": 15360, + "layer_types": [ + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention" + ], + "max_position_embeddings": 131072, + "model_type": "gemma3_text", + "num_attention_heads": 16, + "num_hidden_layers": 48, + "num_key_value_heads": 8, + "query_pre_attn_scalar": 256, + "rms_norm_eps": 1e-06, + "rope_local_base_freq": 10000.0, + "rope_scaling": { + "factor": 8.0, + "rope_type": "linear" + }, + "rope_theta": 1000000.0, + "sliding_window": 1024, + "torch_dtype": "bfloat16", + "use_cache": true, + "vocab_size": 262208 + }, + "torch_dtype": "bfloat16", + "transformers_version": "4.55.4", + "vision_config": { + "attention_dropout": 0.0, + "hidden_act": "gelu_pytorch_tanh", + "hidden_size": 1152, + "image_size": 896, + "intermediate_size": 4304, + "layer_norm_eps": 1e-06, + "model_type": "siglip_vision_model", + "num_attention_heads": 16, + "num_channels": 3, + "num_hidden_layers": 27, + "patch_size": 14, + "torch_dtype": "bfloat16", + "vision_use_head": false + } +}