`disable_progress_bar` is ignored for some models (Loading checkpoint shards)

[JonasLoos]

Describe the bug

When loading some pipelines, diffusers.utils.logging.disable_progress_bar() doesn't disable all progress bars. In particular the "Loading checkpoint shards" progress bar still appears. The "Loading pipeline components..." progress bar, however, is disabled as expected. Models I found, where this occurs, are:

• stabilityai/stable-diffusion-3-medium-diffusers
• black-forest-labs/FLUX.1-schnell

The image generation progress bar also doesn't respect this setting, but can be disabled with pipe.set_progress_bar_config(disable=True). When files are downloaded, the progress bars are also not disabled. These two cases seem like they might be intentional. Are they?

Is there better way to disable progress bars globally for diffusers? Can the "Loading checkpoint shards" progress bar be disabled specifically?

Reproduction

import diffusers
diffusers.utils.logging.disable_progress_bar()
# pipe = diffusers.StableDiffusion3Pipeline.from_pretrained('stabilityai/stable-diffusion-3-medium-diffusers')
pipe = diffusers.FluxPipeline.from_pretrained('black-forest-labs/FLUX.1-schnell')
pipe('test')

Logs

>>> pipe = diffusers.FluxPipeline.from_pretrained('black-forest-labs/FLUX.1-schnell')
Loading checkpoint shards: 100%|██████████████████████████████████████████████████████████| 2/2 [00:03<00:00,  1.56s/it]
You set `add_prefix_space`. The tokenizer needs to be converted from the slow tokenizers
>>>

System Info

Google Colab

or locally:

• 🤗 Diffusers version: 0.30.3
• Running on Google Colab?: No
• Python version: 3.12.7
• PyTorch version (GPU?): 2.5.0+cu124 (True)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Huggingface_hub version: 0.26.0
• Transformers version: 4.45.2
• Accelerate version: 1.0.1
• PEFT version: not installed
• Bitsandbytes version: not installed
• Safetensors version: 0.4.5
• xFormers version: not installed

Who can help?

@sayakpaul @DN6

[JonasLoos]

It seems like the "Loading checkpoint shards" progress bar occurs when the T5EncoderModel is loaded (for flux).

from transformers import T5EncoderModel
T5EncoderModel.from_pretrained('black-forest-labs/FLUX.1-schnell', subfolder="text_encoder_2")

As this submodule uses the transformers library, the issue might be that the disable_progress_bar setting isn't passed on to it.

---

Edit:

Disabling the progress bar for the transformers library actually successfully hides the "Loading checkpoint shards" progress bar.

import transformers
transformers.utils.logging.disable_progress_bar()

This solves the question on how to disable this progress bar. Remaining question:

Is there better way to disable all progress bars for diffusers?

[a-r-r-o-w]

Hi. As you've figured out, this is the intended method for disabling progress bars. Diffusers will disable progress bars relevant to the models/pipelines provided by it, and same goes for transformers. As it's quite simple to do for both libraries, there isn't a need to support something that disables it for both from within diffusers.

Marking this issue as resolved since you've figured out a solution, but feel free to re-open in case there's anything else.