{ "quant_method": "auto-round", "packing_format": "auto_round:llm_compressor", "bits": 8, "group_size": 32, "sym": true, "data_type": "mx_fp", "iters": 0, "model_free": true, "autoround_version": "0.17.0", "enable_quanted_input": false, "block_name_to_quantize": "transformer_blocks,single_transformer_blocks", "act_bits": 8, "act_data_type": "mx_fp", "act_dynamic": true, "act_group_size": 32, "act_sym": true, "extra_config": { "double_stream_modulation_img.linear": { "bits": 16, "data_type": "float", "act_bits": 16, "act_data_type": "float" }, "double_stream_modulation_txt.linear": { "bits": 16, "data_type": "float", "act_bits": 16, "act_data_type": "float" }, "single_stream_modulation.linear": { "bits": 16, "data_type": "float", "act_bits": 16, "act_data_type": "float" }, "norm_out.linear": { "bits": 16, "data_type": "float", "act_bits": 16, "act_data_type": "float" }, "proj_out": { "bits": 16, "data_type": "float", "act_bits": 16, "act_data_type": "float" } } }