Missing binding for cuda.empty_cache

[xd009642]

In the tch and torch-sys crates there doesn't appear to be a version of https://pytorch.org/docs/stable/generated/torch.cuda.empty_cache.html#torch-cuda-empty-cache or torch._C.cuda_emptyCache which it called. I'll have a deeper look into this and PRing it but any guidance would be appreciated as this is a fairly important feature when sharing GPUs with other jobs.

[xd009642]

That and the other memory controls - but for my own issue I'm happy going for the blunt force approach to get torch to free up some of it's excessive allocations.

[xd009642]

Bumping slightly as this is posing more of a problem. I'm probably going to resort to using https://crates.io/crates/nvml-wrapper in the short term to detect issues before they happen and otherwise try to find some time to look into how empty_cache would be implemented