[export][reland] Convert autocast to HOO #132677
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Summary: Reland of D60206382. Suggested in pytorch#128394. If there's an autocast context manager, the predispatch (strict) graph can look something like: ``` class <lambda>(torch.nn.Module): def forward(self, x: "f32[1]"): ... _enter_autocast = torch.amp.autocast_mode._enter_autocast('cuda', torch.bfloat16, True, None) mm: "f32[8, 8]" = torch.ops.aten.mm.default(rand, rand_1); rand = rand_1 = None _exit_autocast = torch.amp.autocast_mode._exit_autocast(_enter_autocast); _enter_autocast = None return (mm_1,) ``` But the operator `torch.amp.autocast_mode._enter_autocast` is not a valid ATen op. We remove these nodes by turning autocast into a higher order operator and make a submodule for the blocks between `_enter_autocast` and `_exit_autocast`. Some potential followup improvement: 1) Merge some of the duplicated logic with `replace_set_grad_with_hop_pass.py` 2) Check the current autocast status (any enabled? dtype?) and not create a submodule if the autocast args matches current autocast status. Test Plan: CI ``` buck2 run 'fbcode//mode/dev-nosan' fbcode//caffe2/test:test_export -- -r "test_predispatch_autocast" buck2 run 'fbcode//mode/dev-nosan' fbcode//caffe2/test:test_export -- -r "test_predispatch_set_grad" ``` Verified that now we can export the llama model in gh issue 128394 and the gemma model in gh issue 131829 without error. Differential Revision: D60770038
|
This pull request was exported from Phabricator. Differential Revision: D60770038 |
angelayi
approved these changes
Aug 5, 2024
|
@pytorchbot merge -f 'Landed internally' (Initiating merge automatically since Phabricator Diff has merged, using force because this PR might not pass merge_rules.json but landed internally) |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
This was referenced Aug 6, 2024
angelayi
pushed a commit
that referenced
this pull request
Aug 7, 2024
Summary: Reland of D60206382. Suggested in #128394. If there's an autocast context manager, the predispatch (strict) graph can look something like: ``` class <lambda>(torch.nn.Module): def forward(self, x: "f32[1]"): ... _enter_autocast = torch.amp.autocast_mode._enter_autocast('cuda', torch.bfloat16, True, None) mm: "f32[8, 8]" = torch.ops.aten.mm.default(rand, rand_1); rand = rand_1 = None _exit_autocast = torch.amp.autocast_mode._exit_autocast(_enter_autocast); _enter_autocast = None return (mm_1,) ``` But the operator `torch.amp.autocast_mode._enter_autocast` is not a valid ATen op. We remove these nodes by turning autocast into a higher order operator and make a submodule for the blocks between `_enter_autocast` and `_exit_autocast`. Some potential followup improvement: 1) Merge some of the duplicated logic with `replace_set_grad_with_hop_pass.py` 2) Check the current autocast status (any enabled? dtype?) and not create a submodule if the autocast args matches current autocast status. Test Plan: CI ``` buck2 run 'fbcode//mode/dev-nosan' fbcode//caffe2/test:test_export -- -r "test_predispatch_autocast" buck2 run 'fbcode//mode/dev-nosan' fbcode//caffe2/test:test_export -- -r "test_predispatch_set_grad" ``` Verified that now we can export the llama model in gh issue 128394 and the gemma model in gh issue 131829 without error. Differential Revision: D60770038 Pull Request resolved: #132677 Approved by: https://github.com/angelayi
This was referenced Aug 7, 2024
angelayi
pushed a commit
that referenced
this pull request
Aug 8, 2024
Summary: Reland of D60206382. Suggested in #128394. If there's an autocast context manager, the predispatch (strict) graph can look something like: ``` class <lambda>(torch.nn.Module): def forward(self, x: "f32[1]"): ... _enter_autocast = torch.amp.autocast_mode._enter_autocast('cuda', torch.bfloat16, True, None) mm: "f32[8, 8]" = torch.ops.aten.mm.default(rand, rand_1); rand = rand_1 = None _exit_autocast = torch.amp.autocast_mode._exit_autocast(_enter_autocast); _enter_autocast = None return (mm_1,) ``` But the operator `torch.amp.autocast_mode._enter_autocast` is not a valid ATen op. We remove these nodes by turning autocast into a higher order operator and make a submodule for the blocks between `_enter_autocast` and `_exit_autocast`. Some potential followup improvement: 1) Merge some of the duplicated logic with `replace_set_grad_with_hop_pass.py` 2) Check the current autocast status (any enabled? dtype?) and not create a submodule if the autocast args matches current autocast status. Test Plan: CI ``` buck2 run 'fbcode//mode/dev-nosan' fbcode//caffe2/test:test_export -- -r "test_predispatch_autocast" buck2 run 'fbcode//mode/dev-nosan' fbcode//caffe2/test:test_export -- -r "test_predispatch_set_grad" ``` Verified that now we can export the llama model in gh issue 128394 and the gemma model in gh issue 131829 without error. Differential Revision: D60770038 Pull Request resolved: #132677 Approved by: https://github.com/angelayi
bluenote10
reviewed
Dec 1, 2024
| def _replace_with_hop_helper( | ||
| node: torch.fx.Node, | ||
| enter_block_node: torch.fx.Node, | ||
| node_filter: Callable, |
There was a problem hiding this comment.
@yushangdi I'm wondering what is the purpose of that node_filter?
As far as I can see it is not used at all, making it a dead argument. Was the idea to use it eventually?
If not, should it be removed? It looks like the usages would become significantly simpler if they don't have to pass in that filter function.
There was a problem hiding this comment.
good catch, yeah I think it can be removed.
@yushangdi I'm wondering what is the purpose if that
node_filter?As far as I can see it is not used at all, making it a dead argument. Was the idea to use it eventually?
If not, should it be removed? It looks like the usages would become significantly simpler if they don't have to pass in that filter function.
There was a problem hiding this comment.
PR for removing it: #141983
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
Reland of D60206382.
Suggested in #128394.
If there's an autocast context manager, the predispatch (strict) graph can look something like:
But the operator
torch.amp.autocast_mode._enter_autocastis not a valid ATen op. We remove these nodes by turning autocast into a higher order operator and make a submodule for the blocks between_enter_autocastand_exit_autocast.Some potential followup improvement:
replace_set_grad_with_hop_pass.pyTest Plan:
CI
Verified that now we can export the llama model in gh issue 128394 and the gemma model in gh issue 131829 without error.
Differential Revision: D60770038