torchrec

torchrec copied to clipboard

fix pyre-extensions to 0.0.27

1

Summary: ATT Reviewed By: colin2328, RenfeiChen-FB Differential Revision: D37160983

YLGH

toggle gradient clipping if set

1

Summary: current setting max_gradients in fused_params does nothing Need to automatically toggle gradient_clipping Differential Revision: D37017267

YLGH

just testing

YLGH

[bugfix] Fix error on empty sharders

4

When passing a empty sharder list to DMP, the `sharder_map` will be a empty dict, which would return `True` for `not sharder_map`. In our experience, it's beneficial to support passing...

zhuzilin

forgot a done in s3 upload script

1

Summary: ATT Differential Revision: D36466766

YLGH

Update Lazy Extension so in sync with Pytorch

1

Summary: Update LazyModuleMixin to be in sync with D36322771 Differential Revision: D36428528

dstaay-fb

Fix sharding context

1

Summary: Properly configure main context to hold multiple child contexts (typically Embedding Collections) Reviewed By: zhangruiskyline, bigning Differential Revision: D36238989

dstaay-fb

Filter out batch_fused from available kernels if fused_params is not explicitly set and contains optimizer

1

Summary: ATT One common mishap with current optimizer fusion is that planner may select batch_fused even if fused_parmas is empty (thus optimizer defaults to SGD). This is dangerous as it...

YLGH

expose ebc interface in ShardedEmbeddingBagCollection

1

Summary: as title Differential Revision: D36149564

colin2328

Add planner support for inference

1

Differential Revision: D35789599

joshuadeng