| # Image + node + Chute for Vocence deploy. Required in the HF repo at build time. | |
| # Keep deps minimal — anything imported by miner.py must not be in the validator's | |
| # banned list (requests/urllib/httpx/aiohttp/socket/huggingface_hub/importlib/torch.hub). | |
| # huggingface_hub is intentionally NOT installed so an accidental future import | |
| # in miner.py fails fast at runtime instead of silently being available. | |
| Image: | |
| from_base: parachutes/base-python:3.12.9 | |
| run_command: | |
| - pip install torch torchaudio transformers accelerate pyyaml soundfile | |
| - pip install -U qwen-tts | |
| set_workdir: /app | |
| NodeSelector: | |
| gpu_count: 1 | |
| min_vram_gb_per_gpu: 24 | |
| include: ["pro_6000"] | |
| exclude: [] | |
| Chute: | |
| tagline: vocence qwen3-tts miner | |
| readme: vocence chute serving qwen3-tts via miner.py (weights pinned in repo) | |
| shutdown_after_seconds: 86400 | |
| concurrency: 1 | |
| max_instances: 1 | |
| scaling_threshold: 0.5 | |
| tee: true | |