Checking download status keeps loading forever

[HamzaMihfad]

After installing Exo on my macbook it was working after downloading DeepSeek 4B model, speed 17tokens per second.

==> but after I tried to install exo on my ubuntu server, everything stoped, exo stopped working on my macbook and the "Checking download status" keeps loading forever

[airdamien]

I am also experiencing this, debug=9 isn't showing anything obviously wrong.

[AI-Guru]

Same here. I did a reinstall into a fresh conda environment. Did not work.

[AlexCheema]

Fixed with #640
We only show models compatible with your hardware now.

[sexyfeifan]

I am also experiencing this

[AlexCheema]

I am also experiencing this

I can't reproduce this issue.

What commit are you on?
Do any errors show up in the console?

[kris330]

set the HF_ENDPOINT environment variable, that will be work!

[veratu]

I have this same issue, here are the steps I followed:

Freshly installed M4 Macbook Pro
Installed Python 3.12
Installed mlx via pip install mlx
Cloned exo git repo
ran: pip install -e . in the exo folder
Everything finished correctly
launch exo
Open web browser to localhost
Get the same error as shown above.

I also tried:

HF_ENDPOINT=https://hf-mirror.com exo

This did not work either.

FIXED:

I ran the Python install Certificates command (a second time) and now it isn't giving me the error above. But models wouldn't download, so then I attempted the HF_ENDPOINT=https://hf-mirror.com exo and it started working. I then just exited and ran exo normally and that worked too.

[wang153723482]

I have this same issue, here are the steps I followed:

Freshly installed M4 Macbook Pro Installed Python 3.12 Installed mlx via pip install mlx Cloned exo git repo ran: pip install -e . in the exo folder Everything finished correctly launch exo Open web browser to localhost Get the same error as shown above.

I also tried:

HF_ENDPOINT=https://hf-mirror.com exo

This did not work either.

FIXED:

I ran the Python install Certificates command (a second time) and now it isn't giving me the error above. But models wouldn't download, so then I attempted the HF_ENDPOINT=https://hf-mirror.com exo and it started working. I then just exited and ran exo normally and that worked too.

It's work.

[PanZongQing]

Is there any other way to do this? I still can't solve the problem by following the above operations.

[MiloWang2048]

This should be the correct log for the issue:

Download error on attempt 1/30 for repo_id='mlx-community/Llama-3.2-1B-Instruct-4bit' revision='main' path='model.safetensors.index.json' target_dir=PosixPath('/var/folders/qr/0l7p6f855ksd6nk3js8fgw9c0000gn/T/exo/mlx-community--Llama-3.2-1B-Instruct-4bit')
Traceback (most recent call last):
  File "/Users/xxx/Desktop/exo/exo/download/new_shard_download.py", line 134, in download_file_with_retry
    try: return await _download_file(repo_id, revision, path, target_dir, on_progress)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/xxx/Desktop/exo/exo/download/new_shard_download.py", line 164, in _download_file
    raise Exception(f"Downloaded file {target_dir/path} has hash {final_hash} but remote hash is {remote_hash}")
Exception: Downloaded file /var/folders/qr/0l7p6f855ksd6nk3js8fgw9c0000gn/T/exo/mlx-community--Llama-3.2-1B-Instruct-4bit/model.safetensors.index.json has hash 32101c2481caabb396a3b36c3fd8b219b0da9c2c but remote hash is 32101c2481caabb396a3b36c3fd8b219b0da9c2c-gzip

Looks like a -gzip postfix triggered unexpwcted checksum mismatch.

[MiloWang2048]

For a quick fix, edit {repo}/exo/download/new_shard_download.py line 160

exo/exo/download/new_shard_download.py

Line 160 in a605e23

integrity = final_hash == remote_hash

from integrity = final_hash == remote_hash to integrity = final_hash in remote_hash

Note: you need to manage your git wktree or reclone to get updates from exo guys.

[sivagaga]

+1 I have the same problem

[AlexCheema]

Thanks for the detailed issue reports -- fixed with #737