Horace He 50cadfae10 Add strictness check and made tensors into leaves if input tensors were leaves (#77474) ... I think this makes sense to do? Otherwise, if you call `backward()` in your traced function, you can't get gradients out of any tensors that should have been leaves. Pull Request resolved: https://github.com/pytorch/pytorch/pull/77474 Approved by: https://github.com/ezyang		2022-05-21 01:16:39 +00:00
..
experimental	Add strictness check and made tensors into leaves if input tensors were leaves (#77474)	2022-05-21 01:16:39 +00:00
passes	Create JSON from new FX IR and lower to LLVM (#77765)	2022-05-20 03:20:57 +00:00
__init__.py	Refactor FX codegen into extensible Codegen object (#72566)	2022-02-11 18:13:29 +00:00
__init__.pyi
_compatibility.py
_pytree.py
_symbolic_trace.py	[Re-submit] Make tracer be able to trace different forward functions	2022-05-17 01:05:33 +00:00
annotate.py
graph_module.py	[WIP] Make constructor calls in experimental MetaTracer serializable	2022-05-11 00:19:47 +00:00
graph.py	[WIP] Make constructor calls in experimental MetaTracer serializable	2022-05-11 00:19:47 +00:00
immutable_collections.py	[FX] Make immutable_list and immutable_dict work with pytrees (#73766)	2022-03-04 19:35:41 +00:00
interpreter.py	[reland] Process inputs and outputs in fx interpreter (#74637)	2022-03-24 18:32:57 +00:00
node.py	record_function: add torchbind alternative API (#72301)	2022-03-08 03:26:27 +00:00
operator_schemas.py	Fix for normalizing signature for op overloads (#77182)	2022-05-10 23:36:26 +00:00
OVERVIEW.md
proxy.py	[Re-submit] Make tracer be able to trace different forward functions	2022-05-17 01:05:33 +00:00
subgraph_rewriter.py	[torch.fx] Fix replace pattern mechanism (#66442)	2021-11-07 13:23:02 -08:00
tensor_type.py