mate-local-models/image/mlx/prism-ml/bonsai-image-ternary-4B-mlx-2bit/transformer-packed-mflux/quantization_config.json

120 lines
4.9 KiB
JSON

{
"format": "mlx-packed-affine",
"solver": "ternary",
"bits": 2,
"group_size": 128,
"scale_dtype": "bfloat16",
"skip_patterns": [
"proj_out",
"x_embedder",
"context_embedder",
"time_text_embed",
"time_guidance_embed",
"norm_out",
"double_stream_modulation_img",
"double_stream_modulation_txt",
"single_stream_modulation"
],
"quantized_modules": [
"single_transformer_blocks.0.attn.to_out",
"single_transformer_blocks.0.attn.to_qkv_mlp_proj",
"single_transformer_blocks.1.attn.to_out",
"single_transformer_blocks.1.attn.to_qkv_mlp_proj",
"single_transformer_blocks.10.attn.to_out",
"single_transformer_blocks.10.attn.to_qkv_mlp_proj",
"single_transformer_blocks.11.attn.to_out",
"single_transformer_blocks.11.attn.to_qkv_mlp_proj",
"single_transformer_blocks.12.attn.to_out",
"single_transformer_blocks.12.attn.to_qkv_mlp_proj",
"single_transformer_blocks.13.attn.to_out",
"single_transformer_blocks.13.attn.to_qkv_mlp_proj",
"single_transformer_blocks.14.attn.to_out",
"single_transformer_blocks.14.attn.to_qkv_mlp_proj",
"single_transformer_blocks.15.attn.to_out",
"single_transformer_blocks.15.attn.to_qkv_mlp_proj",
"single_transformer_blocks.16.attn.to_out",
"single_transformer_blocks.16.attn.to_qkv_mlp_proj",
"single_transformer_blocks.17.attn.to_out",
"single_transformer_blocks.17.attn.to_qkv_mlp_proj",
"single_transformer_blocks.18.attn.to_out",
"single_transformer_blocks.18.attn.to_qkv_mlp_proj",
"single_transformer_blocks.19.attn.to_out",
"single_transformer_blocks.19.attn.to_qkv_mlp_proj",
"single_transformer_blocks.2.attn.to_out",
"single_transformer_blocks.2.attn.to_qkv_mlp_proj",
"single_transformer_blocks.3.attn.to_out",
"single_transformer_blocks.3.attn.to_qkv_mlp_proj",
"single_transformer_blocks.4.attn.to_out",
"single_transformer_blocks.4.attn.to_qkv_mlp_proj",
"single_transformer_blocks.5.attn.to_out",
"single_transformer_blocks.5.attn.to_qkv_mlp_proj",
"single_transformer_blocks.6.attn.to_out",
"single_transformer_blocks.6.attn.to_qkv_mlp_proj",
"single_transformer_blocks.7.attn.to_out",
"single_transformer_blocks.7.attn.to_qkv_mlp_proj",
"single_transformer_blocks.8.attn.to_out",
"single_transformer_blocks.8.attn.to_qkv_mlp_proj",
"single_transformer_blocks.9.attn.to_out",
"single_transformer_blocks.9.attn.to_qkv_mlp_proj",
"transformer_blocks.0.attn.add_k_proj",
"transformer_blocks.0.attn.add_q_proj",
"transformer_blocks.0.attn.add_v_proj",
"transformer_blocks.0.attn.to_add_out",
"transformer_blocks.0.attn.to_k",
"transformer_blocks.0.attn.to_out.0",
"transformer_blocks.0.attn.to_q",
"transformer_blocks.0.attn.to_v",
"transformer_blocks.0.ff.linear_in",
"transformer_blocks.0.ff.linear_out",
"transformer_blocks.0.ff_context.linear_in",
"transformer_blocks.0.ff_context.linear_out",
"transformer_blocks.1.attn.add_k_proj",
"transformer_blocks.1.attn.add_q_proj",
"transformer_blocks.1.attn.add_v_proj",
"transformer_blocks.1.attn.to_add_out",
"transformer_blocks.1.attn.to_k",
"transformer_blocks.1.attn.to_out.0",
"transformer_blocks.1.attn.to_q",
"transformer_blocks.1.attn.to_v",
"transformer_blocks.1.ff.linear_in",
"transformer_blocks.1.ff.linear_out",
"transformer_blocks.1.ff_context.linear_in",
"transformer_blocks.1.ff_context.linear_out",
"transformer_blocks.2.attn.add_k_proj",
"transformer_blocks.2.attn.add_q_proj",
"transformer_blocks.2.attn.add_v_proj",
"transformer_blocks.2.attn.to_add_out",
"transformer_blocks.2.attn.to_k",
"transformer_blocks.2.attn.to_out.0",
"transformer_blocks.2.attn.to_q",
"transformer_blocks.2.attn.to_v",
"transformer_blocks.2.ff.linear_in",
"transformer_blocks.2.ff.linear_out",
"transformer_blocks.2.ff_context.linear_in",
"transformer_blocks.2.ff_context.linear_out",
"transformer_blocks.3.attn.add_k_proj",
"transformer_blocks.3.attn.add_q_proj",
"transformer_blocks.3.attn.add_v_proj",
"transformer_blocks.3.attn.to_add_out",
"transformer_blocks.3.attn.to_k",
"transformer_blocks.3.attn.to_out.0",
"transformer_blocks.3.attn.to_q",
"transformer_blocks.3.attn.to_v",
"transformer_blocks.3.ff.linear_in",
"transformer_blocks.3.ff.linear_out",
"transformer_blocks.3.ff_context.linear_in",
"transformer_blocks.3.ff_context.linear_out",
"transformer_blocks.4.attn.add_k_proj",
"transformer_blocks.4.attn.add_q_proj",
"transformer_blocks.4.attn.add_v_proj",
"transformer_blocks.4.attn.to_add_out",
"transformer_blocks.4.attn.to_k",
"transformer_blocks.4.attn.to_out.0",
"transformer_blocks.4.attn.to_q",
"transformer_blocks.4.attn.to_v",
"transformer_blocks.4.ff.linear_in",
"transformer_blocks.4.ff.linear_out",
"transformer_blocks.4.ff_context.linear_in",
"transformer_blocks.4.ff_context.linear_out"
]
}