Fix: Prevent crashes when HuggingFace is unreachable

[mpecanha]

Voicebox Offline Mode Fix

Problem

Voicebox crashes when generating speech if HuggingFace is unreachable, even when models are fully cached locally.

Root Cause:

• Voicebox downloads mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16 (MLX optimized version)
• But mlx_audio.tts.load() tries to fetch config.json from original repo Qwen/Qwen3-TTS-12Hz-1.7B-Base
• This network request fails → server crashes with RemoteDisconnected

Related Issues:

• Issue Internet connection required, even though models are downloaded? #150: "Internet connection required, even though models are downloaded?"
• Issue API Stability Issues: Model Loading Hangs and Server Crashes #151: "API Stability Issues: Model Loading Hangs and Server Crashes"

Solution

Two-part fix:

1. Monkey-patch huggingface_hub (backend/utils/hf_offline_patch.py)

• Intercepts cache lookup functions
• Forces offline mode early (before mlx_audio imports)
• Adds debug logging for cache hits/misses

2. Symlink original repo to MLX version (ensure_original_qwen_config_cached())

• When original Qwen/Qwen3-TTS-12Hz-1.7B-Base cache doesn't exist
• But MLX mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16 does exist
• Creates a symlink so cache lookups succeed

Files Changed

• backend/backends/mlx_backend.py - Added patch imports at top
• backend/utils/hf_offline_patch.py - New patch module

Testing

To test this fix:

1. Build Voicebox from source: make build
2. Disconnect from internet
3. Try generating speech
4. Should work without network requests

Build Instructions

# Install dependencies
pip install -r requirements.txt

# Build the app
make build

# Or build just the server
make build-server

Notes

• The patch is applied automatically when mlx_backend.py is imported
• Set VOICEBOX_OFFLINE_PATCH=0 to disable the patch
• The symlink approach works because the config.json is compatible between versions

---

Patch contributed by community