Qwen3-ASR-API ghcr.io/hsiang-han/qwen3-asr-api:latest https://ghcr.io bridge bash false https://github.com/hsiang-han/Qwen3-ASR-API/issues https://github.com/hsiang-han/Qwen3-ASR-API Pure OpenAI-compatible Speech-to-Text API powered by Qwen3-ASR via vLLM. State-of-the-art Chinese (22 dialects), English, and 50+ languages. No bloat — just the model and API. Switch between 0.6B (~2-3GB VRAM) and 1.7B (~4-6GB VRAM) via MODEL_ID. CUDA 12.8 (Blackwell/Ada). First start downloads model from HuggingFace. China users: set HF_ENDPOINT to https://hf-mirror.com for faster downloads. Requires NVIDIA GPU with driver 550+. AI: Productivity: Tools: http://[IP]:[PORT:8000]/docs https://raw.githubusercontent.com/hsiang-han/unraid_templates/main/templates/qwen3-asr-api.xml https://raw.githubusercontent.com/hsiang-han/unraid_templates/main/assets/qwen3-asr-api-icon.png --gpus all --shm-size=4g 8000 /mnt/user/appdata/qwen3-asr-api/models Qwen/Qwen3-ASR-0.6B https://huggingface.co 0.8 8192 all compute,utility