search

tenstorrent / tt-metal

:metal: TT-NN operator library, and TT-Metalium low level kernel programming model.
https://docs.tenstorrent.com/ttnn/latest/index.html
Apache License 2.0
485 stars 79 forks source link

Implement auto shard selection for conv2d #13541

Open pavlejosipovic opened 1 month ago

pavlejosipovic commented 1 month ago

In case user passes in interleaved tensor for conv activations, we should automatically select best sharding scheme given input tensors for the convolution. Primary motivation of this work is to reduce number of Out-of-memory issues coming out of pytorch 2.0 ttnn integration.

jvasilje commented 3 weeks ago

This is a P1. @pavlejosipovic this can be your priority, but we should not pollute the P0 bug space.