transformer/quantization_config.json

{
  "add_skip_keys": false,
  "dequantize_fp32": false,
  "group_size": 0,
  "is_integer": true,
  "modules_dtype_dict": {},
  "modules_to_not_convert": [
    ".patch_embedding",
    "norm3",
    "norm1",
    ".condition_embedder",
    ".rope",
    "norm2",
    ".norm_out",
    "scale_shift_table",
    "time_embedder",
    "pos_embed",
    ".proj_out"
  ],
  "non_blocking": false,
  "quant_conv": false,
  "quant_method": "sdnq",
  "quantization_device": null,
  "return_device": null,
  "svd_rank": 32,
  "svd_steps": 8,
  "use_quantized_matmul": false,
  "use_quantized_matmul_conv": false,
  "use_svd": true,
  "weights_dtype": "uint4"
}