Make PyTorch models up to 40% faster! Thunder is a source to source compiler for PyTorch. It enables using different hardware executors at once; across one or thousands of GPUs.
Apache License 2.0
1.14k
stars
73
forks
source link
duplicate symbols from autoregistration (torch.foo vs. torch.Tensor.foo) #1140
Autoregistration registers duplicate symbols under the same name, which is not ideal, and we should go the same way as for manually registered symbols (see below).
Autoregistration registers duplicate symbols under the same name, which is not ideal, and we should go the same way as for manually registered symbols (see below).
Repro:
By contrast, manually registered functions register both with a single symbol: https://github.com/Lightning-AI/lightning-thunder/blob/3770bf02240cdb6ec86a6d5224807bc6ad39d2a9/thunder/torch/__init__.py#L251-L256
(credit for discovery and analysis: @k223kim , thank you)