jinx/ComfyUI
1
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-08-02 23:14:49 +08:00
Code Issues Activity

Support scaled fp8 t5xxl model.

Browse Source
This commit is contained in:
comfyanonymous
2024-10-20 22:27:00 -04:00
parent f9f9faface
commit 83ca891118
6 changed files with 63 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
comfy/sd.py
View File

@@ -432,16 +432,15 @@ def detect_te_model(sd):
return None
def t5xxl_weight_dtype(clip_data):
def t5xxl_detect(clip_data):
weight_name = "encoder.block.23.layer.1.DenseReluDense.wi_1.weight"
dtype_t5 = None
for sd in clip_data:
weight = sd.get(weight_name, None)
if weight is not None:
dtype_t5 = weight.dtype
break
return dtype_t5
if weight_name in sd:
return comfy.text_encoders.sd3_clip.t5_xxl_detect(sd)
return {}
def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip_type=CLIPType.STABLE_DIFFUSION, model_options={}):
@@ -475,7 +474,7 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
clip_target.clip = comfy.text_encoders.sd2_clip.SD2ClipModel
clip_target.tokenizer = comfy.text_encoders.sd2_clip.SD2Tokenizer
elif te_model == TEModel.T5_XXL:
clip_target.clip = comfy.text_encoders.sd3_clip.sd3_clip(clip_l=False, clip_g=False, t5=True, dtype_t5=t5xxl_weight_dtype(clip_data))
clip_target.clip = comfy.text_encoders.sd3_clip.sd3_clip(clip_l=False, clip_g=False, t5=True, **t5xxl_detect(clip_data))
clip_target.tokenizer = comfy.text_encoders.sd3_clip.SD3Tokenizer
elif te_model == TEModel.T5_XL:
clip_target.clip = comfy.text_encoders.aura_t5.AuraT5Model
@@ -493,19 +492,19 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
elif len(clip_data) == 2:
if clip_type == CLIPType.SD3:
te_models = [detect_te_model(clip_data[0]), detect_te_model(clip_data[1])]
clip_target.clip = comfy.text_encoders.sd3_clip.sd3_clip(clip_l=TEModel.CLIP_L in te_models, clip_g=TEModel.CLIP_G in te_models, t5=TEModel.T5_XXL in te_models, dtype_t5=t5xxl_weight_dtype(clip_data))
clip_target.clip = comfy.text_encoders.sd3_clip.sd3_clip(clip_l=TEModel.CLIP_L in te_models, clip_g=TEModel.CLIP_G in te_models, t5=TEModel.T5_XXL in te_models, **t5xxl_detect(clip_data))
clip_target.tokenizer = comfy.text_encoders.sd3_clip.SD3Tokenizer
elif clip_type == CLIPType.HUNYUAN_DIT:
clip_target.clip = comfy.text_encoders.hydit.HyditModel
clip_target.tokenizer = comfy.text_encoders.hydit.HyditTokenizer
elif clip_type == CLIPType.FLUX:
clip_target.clip = comfy.text_encoders.flux.flux_clip(dtype_t5=t5xxl_weight_dtype(clip_data))
clip_target.clip = comfy.text_encoders.flux.flux_clip(**t5xxl_detect(clip_data))
clip_target.tokenizer = comfy.text_encoders.flux.FluxTokenizer
else:
clip_target.clip = sdxl_clip.SDXLClipModel
clip_target.tokenizer = sdxl_clip.SDXLTokenizer
elif len(clip_data) == 3:
clip_target.clip = comfy.text_encoders.sd3_clip.sd3_clip(dtype_t5=t5xxl_weight_dtype(clip_data))
clip_target.clip = comfy.text_encoders.sd3_clip.sd3_clip(**t5xxl_detect(clip_data))
clip_target.tokenizer = comfy.text_encoders.sd3_clip.SD3Tokenizer
parameters = 0