Fix tf in bleurt takes all the memory

CannyLab / vdtk

Visual Description Dataset Analysis Toolkit

MIT License

10 stars 2 forks source link

Fix tf in bleurt takes all the memory #6

Closed xk-huang closed 1 year ago

xk-huang commented 1 year ago

tensorflow occupies all the GPU memory, and in "all" mode, the following torch code will fail with OOM. Thus we need to use a single sub-process.

ref: https://github.com/tensorflow/tensorflow/issues/36465#issuecomment-582749350

xk-huang commented 1 year ago

I‘ve fixed that!

Besides, the external jar files and data stored on git-lfs are not accessible due to the insufficient quota. Maybe they could be hosted on huggingface.