Spaces:

build-small-hackathon
/

aMuseMe

Running on Zero

app.py — Added SPACES_ZERO_GPU env-var detection so spaces.GPU is only applied on actual ZeroGPU Spaces. On a permanent GPU (T4 etc.) inference runs directly without trying to allocate a ZeroGPU session (which raises RuntimeError on non-ZeroGPU hardware). ZeroGPU duration also bumped to 150 s.
requirements.txt — Removed ctranslate2==4.3.1 pin. That wheel was compiled against cuDNN 8 and fails with libcudnn_ops_infer.so.8: cannot open shared object file on ZeroGPU (cuDNN 9). Letting pip resolve the wheel automatically fixes this.
packages.txt (new file) — Adds fonts-dejavu-core and fonts-liberation so the Pillow renderer can locate TTF fonts inside the Space container. Without this the video renders with a fallback bitmap font.
src/amuseme/transcriber.py — Whisper model loader now checks torch.cuda.is_available() upfront instead of defaulting to CUDA and catching the cuDNN error at runtime.

After merging

This Space will run correctly on T4 (or any permanent GPU) without dependency issues — no ZeroGPU required. Set the Space hardware to T4 small/medium and it works out of the box.

Reference working Space (T4): https://huggingface.co/spaces/Blazestorm001/aMuseMe-v1

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

Ready to merge

This branch is ready to get merged automatically.

· Sign up or log in to comment