hfendpoints-images

/

nvidia-nemo-asr

like 0

Follow

Inference Endpoints Images 61

This PR pins the PyTorch version as per https://github.com/NVIDIA/NeMo/issues/14107#issuecomment-3052236112 since with the default PyTorch version installed being PyTorch 2.7.1 there was an issue when running the transcription on CUDA devices due to an issue with the CUDA graphs. Additionally, this PR also fixes the formatting with ruff, and updates the Dockerfile as of the recent push of the Docker images at https://hub.docker.com/r/huggingface/hfendpoints-sdk (and pins the version to v0.2.0, being https://github.com/huggingface/hfendpoints/tree/v0.2.0 with the audio-related features, which were missing in the main branch).

cc @mfuntowicz

Thanks for making all the changes @alvarobartt ! LGTM