jimmycarter
commited on
Commit
•
e5befa2
1
Parent(s):
b8de496
Final pipeline fixes
Browse files- README.md +1 -1
- pipeline.py +8 -8
README.md
CHANGED
@@ -69,12 +69,12 @@ quantize(
|
|
69 |
freeze(pipe.transformer)
|
70 |
pipe.enable_model_cpu_offload()
|
71 |
|
72 |
-
# If you are still running out of memory, add do_batch_cfg=False below.
|
73 |
images = pipe(
|
74 |
prompt=prompt,
|
75 |
negative_prompt=negative_prompt,
|
76 |
device=None,
|
77 |
return_dict=False,
|
|
|
78 |
)
|
79 |
images[0][0].save('chalkboard.png')
|
80 |
```
|
|
|
69 |
freeze(pipe.transformer)
|
70 |
pipe.enable_model_cpu_offload()
|
71 |
|
|
|
72 |
images = pipe(
|
73 |
prompt=prompt,
|
74 |
negative_prompt=negative_prompt,
|
75 |
device=None,
|
76 |
return_dict=False,
|
77 |
+
do_batch_cfg=False, # https://github.com/huggingface/optimum-quanto/issues/327
|
78 |
)
|
79 |
images[0][0].save('chalkboard.png')
|
80 |
```
|
pipeline.py
CHANGED
@@ -1614,14 +1614,14 @@ class CustomPipeline(DiffusionPipeline, SD3LoraLoaderMixin):
|
|
1614 |
if guidance_scale_real > 1.0 and i >= no_cfg_until_timestep:
|
1615 |
progress_bar.set_postfix(
|
1616 |
{
|
1617 |
-
'ts': t / 1000.0,
|
1618 |
'cfg': self._guidance_scale_real,
|
1619 |
},
|
1620 |
)
|
1621 |
else:
|
1622 |
progress_bar.set_postfix(
|
1623 |
{
|
1624 |
-
'ts': t / 1000.0,
|
1625 |
'cfg': 'N/A',
|
1626 |
},
|
1627 |
)
|
@@ -1658,17 +1658,17 @@ class CustomPipeline(DiffusionPipeline, SD3LoraLoaderMixin):
|
|
1658 |
# Prepare extra transformer arguments
|
1659 |
extra_transformer_args = {}
|
1660 |
if prompt_mask is not None:
|
1661 |
-
extra_transformer_args["attention_mask"] = prompt_mask_input.to(device=self.transformer.device)
|
1662 |
|
1663 |
# Forward pass through the transformer
|
1664 |
noise_pred = self.transformer(
|
1665 |
-
hidden_states=latent_model_input.to(device=self.transformer.device)
|
1666 |
timestep=timestep / 1000,
|
1667 |
guidance=guidance,
|
1668 |
-
pooled_projections=pooled_prompt_embeds_input.to(device=self.transformer.device)
|
1669 |
-
encoder_hidden_states=prompt_embeds_input.to(device=self.transformer.device)
|
1670 |
-
txt_ids=text_ids_input.to(device=self.transformer.device)
|
1671 |
-
img_ids=latent_image_ids_input.to(device=self.transformer.device)
|
1672 |
joint_attention_kwargs=self.joint_attention_kwargs,
|
1673 |
return_dict=False,
|
1674 |
**extra_transformer_args,
|
|
|
1614 |
if guidance_scale_real > 1.0 and i >= no_cfg_until_timestep:
|
1615 |
progress_bar.set_postfix(
|
1616 |
{
|
1617 |
+
'ts': t.detach().item() / 1000.0,
|
1618 |
'cfg': self._guidance_scale_real,
|
1619 |
},
|
1620 |
)
|
1621 |
else:
|
1622 |
progress_bar.set_postfix(
|
1623 |
{
|
1624 |
+
'ts': t.detach().item() / 1000.0,
|
1625 |
'cfg': 'N/A',
|
1626 |
},
|
1627 |
)
|
|
|
1658 |
# Prepare extra transformer arguments
|
1659 |
extra_transformer_args = {}
|
1660 |
if prompt_mask is not None:
|
1661 |
+
extra_transformer_args["attention_mask"] = prompt_mask_input.to(device=self.transformer.device)
|
1662 |
|
1663 |
# Forward pass through the transformer
|
1664 |
noise_pred = self.transformer(
|
1665 |
+
hidden_states=latent_model_input.to(device=self.transformer.device),
|
1666 |
timestep=timestep / 1000,
|
1667 |
guidance=guidance,
|
1668 |
+
pooled_projections=pooled_prompt_embeds_input.to(device=self.transformer.device),
|
1669 |
+
encoder_hidden_states=prompt_embeds_input.to(device=self.transformer.device),
|
1670 |
+
txt_ids=text_ids_input.to(device=self.transformer.device) if text_ids is not None else None,
|
1671 |
+
img_ids=latent_image_ids_input.to(device=self.transformer.device) if latent_image_ids is not None else None,
|
1672 |
joint_attention_kwargs=self.joint_attention_kwargs,
|
1673 |
return_dict=False,
|
1674 |
**extra_transformer_args,
|