Spaces:
Running
on
Zero
Running
on
Zero
AlekseyCalvin
commited on
Commit
•
f152e69
1
Parent(s):
01c68ec
Update pipeline.py
Browse files- pipeline.py +0 -15
pipeline.py
CHANGED
@@ -163,7 +163,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
163 |
prompt: Union[str, List[str]],
|
164 |
num_images_per_prompt: int = 1,
|
165 |
device: Optional[torch.device] = None,
|
166 |
-
clip_skip: Optional[int] = None,
|
167 |
):
|
168 |
device = device or self._execution_device
|
169 |
|
@@ -190,11 +189,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
190 |
)
|
191 |
prompt_embeds = self.text_encoder(text_input_ids.to(device), output_hidden_states=False)
|
192 |
|
193 |
-
if clip_skip is None:
|
194 |
-
prompt_embeds = prompt_embeds.hidden_states[-2]
|
195 |
-
else:
|
196 |
-
prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]
|
197 |
-
|
198 |
# Use pooled output of CLIPTextModel
|
199 |
prompt_embeds = prompt_embeds.pooler_output
|
200 |
prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
|
@@ -220,7 +214,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
220 |
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
|
221 |
negative_pooled_prompt_2_embed: Optional[torch.FloatTensor] = None,
|
222 |
negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
|
223 |
-
clip_skip: Optional[int] = None,
|
224 |
max_sequence_length: int = 512,
|
225 |
lora_scale: Optional[float] = None,
|
226 |
):
|
@@ -258,7 +251,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
258 |
prompt=prompt,
|
259 |
device=device,
|
260 |
num_images_per_prompt=num_images_per_prompt,
|
261 |
-
clip_skip=clip_skip,
|
262 |
)
|
263 |
prompt_embeds = self._get_t5_prompt_embeds(
|
264 |
prompt=prompt_2,
|
@@ -293,7 +285,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
293 |
negative_prompt,
|
294 |
device=device,
|
295 |
num_images_per_prompt=num_images_per_prompt,
|
296 |
-
clip_skip=None,
|
297 |
)
|
298 |
negative_clip_prompt_embeds = torch.cat([negative_prompt_embed, negative_prompt_2_embed], dim=-1)
|
299 |
|
@@ -539,7 +530,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
539 |
output_type: Optional[str] = "pil",
|
540 |
return_dict: bool = True,
|
541 |
joint_attention_kwargs: Optional[Dict[str, Any]] = None,
|
542 |
-
clip_skip: Optional[int] = None,
|
543 |
max_sequence_length: int = 300,
|
544 |
**kwargs,
|
545 |
):
|
@@ -562,7 +552,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
562 |
)
|
563 |
|
564 |
self._guidance_scale = guidance_scale
|
565 |
-
self._clip_skip = clip_skip
|
566 |
self._joint_attention_kwargs = joint_attention_kwargs
|
567 |
self._interrupt = False
|
568 |
|
@@ -597,7 +586,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
597 |
pooled_prompt_embeds=pooled_prompt_embeds,
|
598 |
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
599 |
device=device,
|
600 |
-
clip_skip=self.clip_skip,
|
601 |
num_images_per_prompt=num_images_per_prompt,
|
602 |
max_sequence_length=max_sequence_length,
|
603 |
lora_scale=lora_scale,
|
@@ -723,7 +711,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
723 |
output_type: Optional[str] = "pil",
|
724 |
return_dict: bool = True,
|
725 |
joint_attention_kwargs: Optional[Dict[str, Any]] = None,
|
726 |
-
clip_skip: Optional[int] = None,
|
727 |
max_sequence_length: int = 300,
|
728 |
**kwargs,
|
729 |
):
|
@@ -746,7 +733,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
746 |
)
|
747 |
|
748 |
self._guidance_scale = guidance_scale
|
749 |
-
self._clip_skip = clip_skip
|
750 |
self._joint_attention_kwargs = joint_attention_kwargs
|
751 |
self._interrupt = False
|
752 |
|
@@ -781,7 +767,6 @@ class FluxWithCFGPipeline(DiffusionPipeline, FluxLoraLoaderMixin, FromSingleFile
|
|
781 |
pooled_prompt_embeds=pooled_prompt_embeds,
|
782 |
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
783 |
device=device,
|
784 |
-
clip_skip=self.clip_skip,
|
785 |
num_images_per_prompt=num_images_per_prompt,
|
786 |
max_sequence_length=max_sequence_length,
|
787 |
lora_scale=lora_scale,
|
|
|
163 |
prompt: Union[str, List[str]],
|
164 |
num_images_per_prompt: int = 1,
|
165 |
device: Optional[torch.device] = None,
|
|
|
166 |
):
|
167 |
device = device or self._execution_device
|
168 |
|
|
|
189 |
)
|
190 |
prompt_embeds = self.text_encoder(text_input_ids.to(device), output_hidden_states=False)
|
191 |
|
|
|
|
|
|
|
|
|
|
|
192 |
# Use pooled output of CLIPTextModel
|
193 |
prompt_embeds = prompt_embeds.pooler_output
|
194 |
prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
|
|
|
214 |
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
|
215 |
negative_pooled_prompt_2_embed: Optional[torch.FloatTensor] = None,
|
216 |
negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
|
|
|
217 |
max_sequence_length: int = 512,
|
218 |
lora_scale: Optional[float] = None,
|
219 |
):
|
|
|
251 |
prompt=prompt,
|
252 |
device=device,
|
253 |
num_images_per_prompt=num_images_per_prompt,
|
|
|
254 |
)
|
255 |
prompt_embeds = self._get_t5_prompt_embeds(
|
256 |
prompt=prompt_2,
|
|
|
285 |
negative_prompt,
|
286 |
device=device,
|
287 |
num_images_per_prompt=num_images_per_prompt,
|
|
|
288 |
)
|
289 |
negative_clip_prompt_embeds = torch.cat([negative_prompt_embed, negative_prompt_2_embed], dim=-1)
|
290 |
|
|
|
530 |
output_type: Optional[str] = "pil",
|
531 |
return_dict: bool = True,
|
532 |
joint_attention_kwargs: Optional[Dict[str, Any]] = None,
|
|
|
533 |
max_sequence_length: int = 300,
|
534 |
**kwargs,
|
535 |
):
|
|
|
552 |
)
|
553 |
|
554 |
self._guidance_scale = guidance_scale
|
|
|
555 |
self._joint_attention_kwargs = joint_attention_kwargs
|
556 |
self._interrupt = False
|
557 |
|
|
|
586 |
pooled_prompt_embeds=pooled_prompt_embeds,
|
587 |
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
588 |
device=device,
|
|
|
589 |
num_images_per_prompt=num_images_per_prompt,
|
590 |
max_sequence_length=max_sequence_length,
|
591 |
lora_scale=lora_scale,
|
|
|
711 |
output_type: Optional[str] = "pil",
|
712 |
return_dict: bool = True,
|
713 |
joint_attention_kwargs: Optional[Dict[str, Any]] = None,
|
|
|
714 |
max_sequence_length: int = 300,
|
715 |
**kwargs,
|
716 |
):
|
|
|
733 |
)
|
734 |
|
735 |
self._guidance_scale = guidance_scale
|
|
|
736 |
self._joint_attention_kwargs = joint_attention_kwargs
|
737 |
self._interrupt = False
|
738 |
|
|
|
767 |
pooled_prompt_embeds=pooled_prompt_embeds,
|
768 |
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
769 |
device=device,
|
|
|
770 |
num_images_per_prompt=num_images_per_prompt,
|
771 |
max_sequence_length=max_sequence_length,
|
772 |
lora_scale=lora_scale,
|