Root cause: the yolov5 package's torch.hub download of the yolov5 repo
code writes to the Space's persistent disk on every restart, eventually
exceeding the 50GB quota. Redirect all Torch/HF caches to /tmp, pin
CPU-only torch/torchvision (avoids multi-GB CUDA wheels on cpu-basic
hardware), patch torch.load for PyTorch 2.6+'s weights_only default so
the legacy yolov5 checkpoint still loads, and drop the unused
Dockerfile/tator_inference.py/tator dependency left over from an old
pipeline that the gradio-SDK space never actually used. Bump gradio to
6.9.0.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Dear Hugging Face user,
You are receiving this email because your Space FathomNet/MBARI_Monterey_Bay_Benthic is running on an outdated version of Gradio (3.47.1).
This version of Gradio is subject to security vulnerabilities, including the potential leakage of environment variables and secrets, which that have been patched in recent versions.
We strongly suggest you update your Space to the latest Gradio version (4.36.0). You can do so by editing the README.md metadata of your Space, in particular the sdkVersion field in the YAML metadata:
sdkVersion: "4.36.0"
Check out Gradio changelog here.
