NVIDIA/Fuser

Support tensor `rng_seed` and `rng_offset`

Open

#855 opened on 2023年9月8日

GitHub で見る
 (0 comments) (0 reactions) (0 assignees)C++ (80 forks)github user discovery
Priority = P0good first issue

Repository metrics

Stars
 (392 stars)
PR merge metrics
 (PR metrics pending)

説明

Currently only integer RNG states are supported as per https://github.com/NVIDIA/Fuser/blob/e3a0b3ef528a86012d0a5268f3065b9203ed2784/csrc/python_frontend/python_bindings.cpp#L2426-L2473.

I'd like to feature request the support of 0-dim Tensor as rng_seed and rng_offset as well. The dtype would be int64 and the device would be either cpu and cuda.

On top of that, I'd like the function to return a Tensor of integer or an integer that should be used to advance / increment the offset for the next function call.

コントリビューターガイド