Gradio is not auto downloading Hugging Face model weights - auto downloading on Google Colab but not on Windows

[FurkanGozukara]

Describe the bug

The below code automatically downloading necessary model files and weights on Google Colab but not on Windows

Any ideas? How to fix?

import gradio as gr

from diffusers import DiffusionPipeline
import torch

import base64
from io import BytesIO
import os
import gc

from share_btn import community_icon_html, loading_icon_html, share_js

# SDXL code: https://github.com/huggingface/diffusers/pull/3859

model_dir = os.getenv("SDXL_MODEL_DIR")
access_token = "my_token"

if model_dir:
    # Use local model
    model_key_base = os.path.join(model_dir, "stable-diffusion-xl-base-0.9")
    model_key_refiner = os.path.join(model_dir, "stable-diffusion-xl-refiner-0.9")
else:
    model_key_base = "nichijoufan777/stable-diffusion-xl-base-0.9"
    model_key_refiner = "nichijoufan777/stable-diffusion-xl-refiner-0.9"

# Use refiner (enabled by default)
enable_refiner = os.getenv("ENABLE_REFINER", "true").lower() == "true"
# Output images before the refiner and after the refiner
output_images_before_refiner = os.getenv("OUTPUT_IMAGES_BEFORE_REFINER", "false").lower() == "true"

# Create public link
share = os.getenv("SHARE", "false").lower() == "true"

print("Loading model", model_key_base)
pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_safetensors=True, variant="fp16", use_auth_token=access_token)

pipe.enable_model_cpu_offload()
# pipe.to("cuda")

# if using torch < 2.0
# pipe.enable_xformers_memory_efficient_attention()

# pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)

if enable_refiner:
    print("Loading model", model_key_refiner)
    pipe_refiner = DiffusionPipeline.from_pretrained(model_key_refiner, torch_dtype=torch.float16, use_safetensors=True, variant="fp16", use_auth_token=access_token)
    pipe_refiner.enable_model_cpu_offload()
    # pipe_refiner.to("cuda")

    # if using torch < 2.0
    # pipe_refiner.enable_xformers_memory_efficient_attention()

    # pipe_refiner.unet = torch.compile(pipe_refiner.unet, mode="reduce-overhead", fullgraph=True)

# NOTE: we do not have word list filtering in this gradio demo

is_gpu_busy = False
.
.
.

Reproduction

above

Logs

Loading model nichijoufan777/stable-diffusion-xl-base-0.9
vae\diffusion_pytorch_model.fp16.safetensors not found
Traceback (most recent call last):
  File "C:\stable-diffusion-xl-demo\app.py", line 35, in <module>
    pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_safetensors=True, variant="fp16", use_auth_token=access_token)
  File "C:\stable-diffusion-xl-demo\venv\lib\site-packages\diffusers\pipelines\pipeline_utils.py", line 882, in from_pretrained
    cached_folder = cls.download(
  File "C:\stable-diffusion-xl-demo\venv\lib\site-packages\diffusers\pipelines\pipeline_utils.py", line 1260, in download
    raise EnvironmentError(
OSError: Could not found the necessary `safetensors` weights in {'unet/diffusion_pytorch_model.fp16.safetensors', 'vae/diffusion_pytorch_model.fp16.bin', 'unet/diffusion_pytorch_model.fp16.bin', 'text_encoder/model.fp16.safetensors', 'text_encoder/pytorch_model.fp16.bin', 'text_encoder_2/model.fp16.safetensors', 'vae/diffusion_pytorch_model.fp16.safetensors', 'text_encoder_2/pytorch_model.fp16.bin'} (variant=fp16)
Press any key to continue . . .

System Info

Windows 10, Python 3.10.9

Who can help?

@patrickvonplaten @sayakpaul @will

[will]

I’m sorry I don’t think I can actually help On Jul 6, 2023, at 13:13, Furkan Gözükara ***@***.***> wrote: Describe the bug The below code automatically downloading necessary model files and weights on Google Colab but not on Windows Any ideas? How to fix? import gradio as gr from diffusers import DiffusionPipeline import torch import base64 from io import BytesIO import os import gc from share_btn import community_icon_html, loading_icon_html, share_js # SDXL code: #3859 model_dir = os.getenv("SDXL_MODEL_DIR") access_token = os.getenv("my_token") if model_dir: # Use local model model_key_base = os.path.join(model_dir, "stable-diffusion-xl-base-0.9") model_key_refiner = os.path.join(model_dir, "stable-diffusion-xl-refiner-0.9") else: model_key_base = "nichijoufan777/stable-diffusion-xl-base-0.9" model_key_refiner = "nichijoufan777/stable-diffusion-xl-refiner-0.9" # Use refiner (enabled by default) enable_refiner = os.getenv("ENABLE_REFINER", "true").lower() == "true" # Output images before the refiner and after the refiner output_images_before_refiner = os.getenv("OUTPUT_IMAGES_BEFORE_REFINER", "false").lower() == "true" # Create public link share = os.getenv("SHARE", "false").lower() == "true" print("Loading model", model_key_base) pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_safetensors=True, variant="fp16", use_auth_token=access_token) pipe.enable_model_cpu_offload() # pipe.to("cuda") # if using torch < 2.0 # pipe.enable_xformers_memory_efficient_attention() # pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True) if enable_refiner: print("Loading model", model_key_refiner) pipe_refiner = DiffusionPipeline.from_pretrained(model_key_refiner, torch_dtype=torch.float16, use_safetensors=True, variant="fp16", use_auth_token=access_token) pipe_refiner.enable_model_cpu_offload() # pipe_refiner.to("cuda") # if using torch < 2.0 # pipe_refiner.enable_xformers_memory_efficient_attention() # pipe_refiner.unet = torch.compile(pipe_refiner.unet, mode="reduce-overhead", fullgraph=True) # NOTE: we do not have word list filtering in this gradio demo is_gpu_busy = False . . . Reproduction above Logs No response System Info Windows 10, Python 3.10.9 Who can help? @patrickvonplaten @sayakpaul @will —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: ***@***.***>

[patrickvonplaten]

Just to better understand can you try if any of the following work on Windows:
1.

pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_safetensors=True, use_auth_token=access_token)

or

pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_auth_token=access_token)

[FurkanGozukara]

Just to better understand can you try if any of the following work on Windows: 1.

pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_safetensors=True, use_auth_token=access_token)

or

pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_auth_token=access_token)

this one started downloading ty so much testing

pipe = DiffusionPipeline.from_pretrained(model_key_base, torch_dtype=torch.float16, use_auth_token=access_token)

[patrickvonplaten]

Does it download safetensor weights or PyTorch ones? Not a big problem as both will work, but just curious

[FurkanGozukara]

Does it download safetensor weights or PyTorch ones? Not a big problem as both will work, but just curious

deleted cache folder and started again

here status looking correct?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.