Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-06 13:00:44 +02:00
Code Issues Packages Projects Releases Wiki Activity

cleanup

Browse Source
This commit is contained in:
Disty0
2025-10-08 00:41:04 +03:00
parent df03ea9ba8
commit 7fdf400e8b
2 changed files with 16 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sdnq/loader.py
+1 -1
View File
@@ -66,7 +66,7 @@ def load_sdnq_model(model_path: str, model_cls: ModelMixin = None, file_name: st
model = model_cls._from_config(config) # pylint: disable=protected-access
else:
raise ValueError(f"Dont know how to load model for {model_cls}")
model = sdnq_post_load_quant(model, **quantization_config)
model = sdnq_post_load_quant(model, add_skip_keys=False, **quantization_config)
state_dict = {}
if file_name:
modules/sdnq/quantizer.py
+15 -14
View File
@@ -321,26 +321,27 @@ def apply_sdnq_to_module(model, weights_dtype="int8", torch_dtype=None, group_si
return model
def sdnq_post_load_quant(model, weights_dtype="int8", torch_dtype=None, group_size=0, svd_rank=32, use_svd=False, quant_conv=False, use_quantized_matmul=False, use_quantized_matmul_conv=False, dequantize_fp32=False, non_blocking=False, quantization_device=None, return_device=None, modules_to_not_convert: List[str] = None, modules_dtype_dict: Dict[str, List[str]] = None, op=None): # pylint: disable=unused-argument
model.eval()
def sdnq_post_load_quant(model, weights_dtype="int8", torch_dtype=None, group_size=0, svd_rank=32, use_svd=False, quant_conv=False, use_quantized_matmul=False, use_quantized_matmul_conv=False, dequantize_fp32=False, non_blocking=False, quantization_device=None, return_device=None, modules_to_not_convert: List[str] = None, modules_dtype_dict: Dict[str, List[str]] = None, add_skip_keys=True, op=None): # pylint: disable=unused-argument
if modules_to_not_convert is None:
modules_to_not_convert = []
if modules_dtype_dict is None:
modules_dtype_dict = {}
if getattr(model, "_keep_in_fp32_modules", None) is not None:
modules_to_not_convert.extend(model._keep_in_fp32_modules) # pylint: disable=protected-access
if getattr(model, "_skip_layerwise_casting_patterns", None) is not None:
modules_to_not_convert.extend(model._skip_layerwise_casting_patterns) # pylint: disable=protected-access
if model.__class__.__name__ == "ChromaTransformer2DModel":
modules_to_not_convert.append("distilled_guidance_layer")
elif model.__class__.__name__ == "QwenImageTransformer2DModel":
modules_to_not_convert.extend(["transformer_blocks.0.img_mod.1.weight", "time_text_embed", "img_in", "txt_in", "proj_out", "norm_out", "pos_embed"])
if "minimum_6bit" not in modules_dtype_dict.keys():
modules_dtype_dict["minimum_6bit"] = ["img_mod"]
else:
modules_dtype_dict["minimum_6bit"].append("img_mod")
if add_skip_keys:
if getattr(model, "_keep_in_fp32_modules", None) is not None:
modules_to_not_convert.extend(model._keep_in_fp32_modules) # pylint: disable=protected-access
if getattr(model, "_skip_layerwise_casting_patterns", None) is not None:
modules_to_not_convert.extend(model._skip_layerwise_casting_patterns) # pylint: disable=protected-access
if model.__class__.__name__ == "ChromaTransformer2DModel":
modules_to_not_convert.append("distilled_guidance_layer")
elif model.__class__.__name__ == "QwenImageTransformer2DModel":
modules_to_not_convert.extend(["transformer_blocks.0.img_mod.1.weight", "time_text_embed", "img_in", "txt_in", "proj_out", "norm_out", "pos_embed"])
if "minimum_6bit" not in modules_dtype_dict.keys():
modules_dtype_dict["minimum_6bit"] = ["img_mod"]
else:
modules_dtype_dict["minimum_6bit"].append("img_mod")
model.eval()
model = apply_sdnq_to_module(
model,
weights_dtype=weights_dtype,