Post

Confirmed working, but when you run for the first time, it starts downloading, but doesn't display the download status, not in the terminal, not in the UI, so it looks like it's stuck at 0% for minutes.

Looks like it's because it uses curl to download the model.

Couldn't we use huggingface_hub with gradio, which should be able to display the full progress so the user doesn't get confused?

It is running. The UI looks frozen because the downloader prints one line and then goes silent.

What I found:

• The app itself is up (start.js running, UI at http://127.0.0.1:42883).
• First generation kicked off an on-demand download of the 8-bit weights. Install never downloads the model; that happens the first time you generate.
• Active process: curl fetching https://huggingface.co/ahmadw/YuE2-3B-MLX/resolve/main/8bit/model.safetensors
• Progress on disk: ~4.15 GB of 4.26 GB (~97%). Remaining is about 110 MB. At the rate I saw, it should finish in under a minute, then it still has to pull a few small files (vae.safetensors, configs, tokenizer) and load the model.

Why it looks stuck:

1. Gradio stays at “Loading MLX model…” 0% the whole time. Download progress is never sent to that bar.
2. The download uses curl -fsSL with output captured, so Pinokio’s terminal never gets a progress bar. The last log line is just [model] downloading model.safetensors... until curl exits.
3. The log file hasn’t been updated since 11:55 for the same reason: Python is blocked on subprocess.run until the 4.2 GB file finishes.

So this is not a hung process. It’s a ~4.2 GB Hugging Face download with no progress reporting.

Don’t stop the app or click Generate again while this is going. If you interrupt it, a partial model.safetensors is left on disk, and the next run will treat that incomplete file as “already downloaded” and skip the fetch.

After this file finishes you should see [model] download complete then [model] loading in the terminal. If it sits past that with no file growth, say so and I’ll dig into the next step.