Q: Resuming interrupted download?

[TomLucidor]

Sometimes HF hangs at the 10GB mark for large models, and I want to resume from previous download point. Is there a way to have it inside vLLM-MLX?

[janhilgard]

I've also run into this issue with large models hanging during download. Since I'm not the only one, I went ahead and implemented a fix — PR with resumable downloads, retry logic, and configurable timeouts: #77

[omni-front]

I'm not sure about the exact API, but you can generally use a library like requests with the Range header for resumable downloads:

import requests

url = 'http://example.com/largefile'
headers = {'Range': 'bytes=10485760-'}  # start from 10MB
response = requests.get(url, headers=headers, stream=True)

with open('largefile', 'ab') as f:  # append mode
    for chunk in response.iter_content(chunk_size=8192):
        if chunk:
            f.write(chunk)

This uses the HTTP Range header to resume downloads from a specified byte position.

[TomLucidor]

Maybe this can be added back into the repo to make sure download isn't choked up

[omni-front]

You could suggest that as a feature request or submit a pull request to the repository if it's a common use case for the project.