jaidaken/ComfyUI
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Make old scaled fp8 format use the new mixed quant ops system. (#11000)

Browse Source
This commit is contained in:
comfyanonymous
2025-12-05 11:35:42 -08:00
committed by GitHub
parent 0ec05b1481
commit 43071e3de3
24 changed files with 278 additions and 275 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
comfy/text_encoders/hunyuan_image.py
View File

@@ -40,10 +40,10 @@ class HunyuanImageTokenizer(QwenImageTokenizer):
class Qwen25_7BVLIModel(sd1_clip.SDClipModel):
def __init__(self, device="cpu", layer="hidden", layer_idx=-3, dtype=None, attention_mask=True, model_options={}):
llama_scaled_fp8 = model_options.get("qwen_scaled_fp8", None)
if llama_scaled_fp8 is not None:
llama_quantization_metadata = model_options.get("llama_quantization_metadata", None)
if llama_quantization_metadata is not None:
model_options = model_options.copy()
model_options["scaled_fp8"] = llama_scaled_fp8
model_options["quantization_metadata"] = llama_quantization_metadata
super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"pad": 151643}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Qwen25_7BVLI, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
@@ -91,12 +91,12 @@ class HunyuanImageTEModel(QwenImageTEModel):
else:
return super().load_sd(sd)
def te(byt5=True, dtype_llama=None, llama_scaled_fp8=None):
def te(byt5=True, dtype_llama=None, llama_quantization_metadata=None):
class QwenImageTEModel_(HunyuanImageTEModel):
def __init__(self, device="cpu", dtype=None, model_options={}):
if llama_scaled_fp8 is not None and "scaled_fp8" not in model_options:
if llama_quantization_metadata is not None:
model_options = model_options.copy()
model_options["qwen_scaled_fp8"] = llama_scaled_fp8
model_options["llama_quantization_metadata"] = llama_quantization_metadata
if dtype_llama is not None:
dtype = dtype_llama
super().__init__(byt5=byt5, device=device, dtype=dtype, model_options=model_options)