Skip to content

[From single file] Make accelerate optional #4132

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jul 17, 2023
Merged

[From single file] Make accelerate optional #4132

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
020eef7
Make accelerate optional
patrickvonplaten Jul 17, 2023
e95170a
make accelerate optional
patrickvonplaten Jul 17, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
50 changes: 31 additions & 19 deletions src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
""" Conversion script for the Stable Diffusion checkpoints."""

import re
from contextlib import nullcontext
from io import BytesIO
from typing import Optional

Expand Down Expand Up @@ -779,7 +780,8 @@ def convert_ldm_clip_checkpoint(checkpoint, local_files_only=False, text_encoder
config_name = "openai/clip-vit-large-patch14"
config = CLIPTextConfig.from_pretrained(config_name)

with init_empty_weights():
ctx = init_empty_weights if is_accelerate_available() else nullcontext
with ctx():
text_model = CLIPTextModel(config)

keys = list(checkpoint.keys())
Expand All @@ -793,8 +795,11 @@ def convert_ldm_clip_checkpoint(checkpoint, local_files_only=False, text_encoder
if key.startswith(prefix):
text_model_dict[key[len(prefix + ".") :]] = checkpoint[key]

for param_name, param in text_model_dict.items():
set_module_tensor_to_device(text_model, param_name, "cpu", value=param)
if is_accelerate_available():
for param_name, param in text_model_dict.items():
set_module_tensor_to_device(text_model, param_name, "cpu", value=param)
else:
text_model.load_state_dict(text_model_dict)

return text_model

Expand Down Expand Up @@ -900,7 +905,8 @@ def convert_open_clip_checkpoint(
# )
config = CLIPTextConfig.from_pretrained(config_name, **config_kwargs)

with init_empty_weights():
ctx = init_empty_weights if is_accelerate_available() else nullcontext
with ctx():
text_model = CLIPTextModelWithProjection(config) if has_projection else CLIPTextModel(config)

keys = list(checkpoint.keys())
Expand Down Expand Up @@ -950,8 +956,11 @@ def convert_open_clip_checkpoint(

text_model_dict[new_key] = checkpoint[key]

for param_name, param in text_model_dict.items():
set_module_tensor_to_device(text_model, param_name, "cpu", value=param)
if is_accelerate_available():
for param_name, param in text_model_dict.items():
set_module_tensor_to_device(text_model, param_name, "cpu", value=param)
else:
text_model.load_state_dict(text_model_dict)

return text_model

Expand Down Expand Up @@ -1172,11 +1181,6 @@ def download_from_original_stable_diffusion_ckpt(
StableUnCLIPPipeline,
)

if not is_accelerate_available():
raise ImportError(
"To correctly use `from_single_file`, please make sure that `accelerate` is installed. You can install it with `pip install accelerate`."
)

if pipeline_class is None:
pipeline_class = StableDiffusionPipeline

Expand Down Expand Up @@ -1346,15 +1350,19 @@ def download_from_original_stable_diffusion_ckpt(
# Convert the UNet2DConditionModel model.
unet_config = create_unet_diffusers_config(original_config, image_size=image_size)
unet_config["upcast_attention"] = upcast_attention
with init_empty_weights():
unet = UNet2DConditionModel(**unet_config)

converted_unet_checkpoint = convert_ldm_unet_checkpoint(
checkpoint, unet_config, path=checkpoint_path, extract_ema=extract_ema
)

for param_name, param in converted_unet_checkpoint.items():
set_module_tensor_to_device(unet, param_name, "cpu", value=param)
ctx = init_empty_weights if is_accelerate_available() else nullcontext
with ctx():
unet = UNet2DConditionModel(**unet_config)

if is_accelerate_available():
for param_name, param in converted_unet_checkpoint.items():
set_module_tensor_to_device(unet, param_name, "cpu", value=param)
else:
unet.load_state_dict(converted_unet_checkpoint)

# Convert the VAE model.
if vae_path is None:
Expand All @@ -1372,11 +1380,15 @@ def download_from_original_stable_diffusion_ckpt(

vae_config["scaling_factor"] = vae_scaling_factor

with init_empty_weights():
ctx = init_empty_weights if is_accelerate_available() else nullcontext
with ctx():
vae = AutoencoderKL(**vae_config)

for param_name, param in converted_vae_checkpoint.items():
set_module_tensor_to_device(vae, param_name, "cpu", value=param)
if is_accelerate_available():
for param_name, param in converted_vae_checkpoint.items():
set_module_tensor_to_device(vae, param_name, "cpu", value=param)
else:
vae.load_state_dict(converted_vae_checkpoint)
else:
vae = AutoencoderKL.from_pretrained(vae_path)

Expand Down