[hoo] Invoke subgraph + effect by angelayi · Pull Request #167231 · pytorch/pytorch

angelayi · 2025-11-06T17:16:03Z

This PR adds support for effectful ops within invoke_subgraphs.

Most of the logic is in invoke_subgraph.py_functionalize_impl.
- In the functionalization metadata collection phase, we note the tokens before going further down the dispatcher, and then note the tokens after coming back from the dispatcher. If there are nodes in the invoke_subgraph subgraph that contain effects, the number of effects should change, or the tokens used for an effect should.
- We will store this effect difference in the InvokeSubgraphCache where the key is the identifier and value is the effect. For now we only support one effect within a subgraph.
- During the tracing part of AOTAutograd, we will then wrap the subgraph to take in and output a token.

Before:

def forward(self, x):
    repeated_subgraph0 = self.repeated_subgraph0
    invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', x)
    return invoke_subgraph

def repeated_subgraph(self, x):
    record_memory = torch.ops.mylib.record_memory.default("forward", "N")
    add = torch.ops.aten.add(x, x)
    return add

After:

def forward(self, token, x):
    repeated_subgraph0 = self.repeated_subgraph0
    invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x)
    getitem = invoke_subgraph[0]  # output token
    getitem_1 = invoke_subgraph[1]
    return (getitem, getitem_1)

def repeated_subgraph(self, token, x):
    with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N')
    getitem = with_effects[0]  # output token
    add = torch.ops.aten.add(x, x)
    return  (getitem, add)

Then there is a bunch of logic within _remove_effect_tokens to handle removing the effects from the invoke_subgraph subgraph

Stack from ghstack (oldest at bottom):

cc @ezyang @EikanWang @jgong5 @wenzhe-nrv

Differential Revision: D87392741

[ghstack-poisoned]

pytorch-bot · 2025-11-06T17:16:09Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/167231

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

ghstack-source-id: 0df0ad4 Pull Request resolved: #167231

cc ezyang EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]

This PR adds support for effectful ops within invoke_subgraphs. * Most of the logic is in `invoke_subgraph.py_functionalize_impl`. * In the functionalization metadata collection phase, we note the tokens before going further down the dispatcher, and then note the tokens after coming back from the dispatcher. If there are nodes in the invoke_subgraph subgraph that contain effects, the number of effects should change, or the tokens used for an effect should. * We will store this effect difference in the `InvokeSubgraphCache` where the key is the identifier and value is the effect. For now we only support one effect within a subgraph. * During the tracing part of AOTAutograd, we will then wrap the subgraph to take in and output a token. Before: ``` def forward(self, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', x) return invoke_subgraph def repeated_subgraph(self, x): record_memory = torch.ops.mylib.record_memory.default("forward", "N") add = torch.ops.aten.add(x, x) return add ``` After: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` * Then there is a bunch of logic within `_remove_effect_tokens` to handle removing the effects from the invoke_subgraph subgraph cc ezyang EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]

anijain2305 · 2025-11-13T22:39:27Z

torch/_higher_order_ops/invoke_subgraph.py

        assert all(
            isinstance(o, (torch.Tensor, int, torch.SymInt, torch.Generator))
            for o in operands
+            if o is not None


when do you see None as input?

The effect tokens are passed in as None here since we will eventually just discard these inputs.

This PR adds support for effectful ops within invoke_subgraphs. * Most of the logic is in `invoke_subgraph.py_functionalize_impl`. * In the functionalization metadata collection phase, we note the tokens before going further down the dispatcher, and then note the tokens after coming back from the dispatcher. If there are nodes in the invoke_subgraph subgraph that contain effects, the number of effects should change, or the tokens used for an effect should. * We will store this effect difference in the `InvokeSubgraphCache` where the key is the identifier and value is the effect. For now we only support one effect within a subgraph. * During the tracing part of AOTAutograd, we will then wrap the subgraph to take in and output a token. Before: ``` def forward(self, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', x) return invoke_subgraph def repeated_subgraph(self, x): record_memory = torch.ops.mylib.record_memory.default("forward", "N") add = torch.ops.aten.add(x, x) return add ``` After: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` * Then there is a bunch of logic within `_remove_effect_tokens` to handle removing the effects from the invoke_subgraph subgraph cc ezyang EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]

pytorchmergebot · 2025-11-20T00:23:56Z

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

This reverts commit f49833d. Reverted #167231 on behalf of https://github.com/yangw-dev due to the diff breaks tests internally ([comment](#167231 (comment)))

pytorchmergebot · 2025-11-20T00:24:09Z

@angelayi your PR has been successfully reverted.

This PR adds support for effectful ops within invoke_subgraphs. * Most of the logic is in `invoke_subgraph.py_functionalize_impl`. * In the functionalization metadata collection phase, we note the tokens before going further down the dispatcher, and then note the tokens after coming back from the dispatcher. If there are nodes in the invoke_subgraph subgraph that contain effects, the number of effects should change, or the tokens used for an effect should. * We will store this effect difference in the `InvokeSubgraphCache` where the key is the identifier and value is the effect. For now we only support one effect within a subgraph. * During the tracing part of AOTAutograd, we will then wrap the subgraph to take in and output a token. Before: ``` def forward(self, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', x) return invoke_subgraph def repeated_subgraph(self, x): record_memory = torch.ops.mylib.record_memory.default("forward", "N") add = torch.ops.aten.add(x, x) return add ``` After: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` * Then there is a bunch of logic within `_remove_effect_tokens` to handle removing the effects from the invoke_subgraph subgraph cc ezyang EikanWang jgong5 wenzhe-nrv Differential Revision: [D87392741](https://our.internmc.facebook.com/intern/diff/D87392741) [ghstack-poisoned]

angelayi · 2025-11-21T21:04:23Z

@angelayi has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

angelayi · 2025-11-21T23:23:25Z

@angelayi has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

pytorchmergebot · 2025-11-24T23:34:21Z

Starting merge as part of PR stack under #167363

…67245) In the [previous PR](https://github.com/pytorch/pytorch/pull/167231/files#diff-e2b74af5d8b538a7d07d18507d27010703742ddad5f819992b55f5abc6d9a502R964-R966) we found that the autograd eager impl of invoke_subgraph calls the subgraph twice. If the subgraph contains effects then effects will be run twice, which is bad. This PR fixes the issue by getting the output metadata from `subgraph`'s `node.meta` if it exists. Differential Revision: [D87392740](https://our.internmc.facebook.com/intern/diff/D87392740) Pull Request resolved: #167245 Approved by: https://github.com/anijain2305 ghstack dependencies: #167231

Updates the implementation of `unlift_tokens` to handle unlifting invoke_subgraph. The context of `unlift_tokens` is currently tokens are threaded as inputs and outputs of the toplevel graph produced by AOTAutograd. However we don't want the inductor traced graph to have any notion of effects/tokens, just that the tokens should introduce some extra dependency behavior. So, we unlift the tokens from the toplevel graph. Instead of placeholder nodes the tokens will come from a `_make_token` call, and instead of outputting the tokens we will sink all tokens into `_sink_tokens`. Similarly, we want the invoke_subgraph subgraph to not have any notion of tokens, so we will also remove the tokens from the inputs of the invoke_subgraph subgraph. However, we still need a way mark the invoke_subgraph call as being effectful at the toplevel module to prevent invoke_subgraph calls from being reordered, so I wrap the invoke_subgraph with an effects. Before: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` After: ``` def forward(self, x): token = torch.ops.prims._make_token.default() repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.with_effects( token, torch.ops.higher_order.invoke_subgraph, repeated_subgraph0, 'subgraph_0', token, x ) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] _ = torch.ops.prims._sink_tokens.default([getitem]) return (getitem_1,) def repeated_subgraph(self, x): token = torch.ops.prims._make_token.default() with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) _ = torch.ops.prims._sink_tokens.default([getitem]) return (add,) ``` Differential Revision: [D87668981](https://our.internmc.facebook.com/intern/diff/D87668981) Pull Request resolved: #167363 Approved by: https://github.com/fxdawnn ghstack dependencies: #167231, #167245

This PR adds support for effectful ops within invoke_subgraphs. * Most of the logic is in `invoke_subgraph.py_functionalize_impl`. * In the functionalization metadata collection phase, we note the tokens before going further down the dispatcher, and then note the tokens after coming back from the dispatcher. If there are nodes in the invoke_subgraph subgraph that contain effects, the number of effects should change, or the tokens used for an effect should. * We will store this effect difference in the `InvokeSubgraphCache` where the key is the identifier and value is the effect. For now we only support one effect within a subgraph. * During the tracing part of AOTAutograd, we will then wrap the subgraph to take in and output a token. Before: ``` def forward(self, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', x) return invoke_subgraph def repeated_subgraph(self, x): record_memory = torch.ops.mylib.record_memory.default("forward", "N") add = torch.ops.aten.add(x, x) return add ``` After: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` * Then there is a bunch of logic within `_remove_effect_tokens` to handle removing the effects from the invoke_subgraph subgraph Differential Revision: [D87392741](https://our.internmc.facebook.com/intern/diff/D87392741) Pull Request resolved: pytorch#167231 Approved by: https://github.com/anijain2305

…torch#167245) In the [previous PR](https://github.com/pytorch/pytorch/pull/167231/files#diff-e2b74af5d8b538a7d07d18507d27010703742ddad5f819992b55f5abc6d9a502R964-R966) we found that the autograd eager impl of invoke_subgraph calls the subgraph twice. If the subgraph contains effects then effects will be run twice, which is bad. This PR fixes the issue by getting the output metadata from `subgraph`'s `node.meta` if it exists. Differential Revision: [D87392740](https://our.internmc.facebook.com/intern/diff/D87392740) Pull Request resolved: pytorch#167245 Approved by: https://github.com/anijain2305 ghstack dependencies: pytorch#167231

Updates the implementation of `unlift_tokens` to handle unlifting invoke_subgraph. The context of `unlift_tokens` is currently tokens are threaded as inputs and outputs of the toplevel graph produced by AOTAutograd. However we don't want the inductor traced graph to have any notion of effects/tokens, just that the tokens should introduce some extra dependency behavior. So, we unlift the tokens from the toplevel graph. Instead of placeholder nodes the tokens will come from a `_make_token` call, and instead of outputting the tokens we will sink all tokens into `_sink_tokens`. Similarly, we want the invoke_subgraph subgraph to not have any notion of tokens, so we will also remove the tokens from the inputs of the invoke_subgraph subgraph. However, we still need a way mark the invoke_subgraph call as being effectful at the toplevel module to prevent invoke_subgraph calls from being reordered, so I wrap the invoke_subgraph with an effects. Before: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` After: ``` def forward(self, x): token = torch.ops.prims._make_token.default() repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.with_effects( token, torch.ops.higher_order.invoke_subgraph, repeated_subgraph0, 'subgraph_0', token, x ) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] _ = torch.ops.prims._sink_tokens.default([getitem]) return (getitem_1,) def repeated_subgraph(self, x): token = torch.ops.prims._make_token.default() with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) _ = torch.ops.prims._sink_tokens.default([getitem]) return (add,) ``` Differential Revision: [D87668981](https://our.internmc.facebook.com/intern/diff/D87668981) Pull Request resolved: pytorch#167363 Approved by: https://github.com/fxdawnn ghstack dependencies: pytorch#167231, pytorch#167245

This reverts commit f49833d. Reverted #167231 on behalf of https://github.com/yangw-dev due to the diff breaks tests internally ([comment](#167231 (comment)))

This PR adds support for effectful ops within invoke_subgraphs. * Most of the logic is in `invoke_subgraph.py_functionalize_impl`. * In the functionalization metadata collection phase, we note the tokens before going further down the dispatcher, and then note the tokens after coming back from the dispatcher. If there are nodes in the invoke_subgraph subgraph that contain effects, the number of effects should change, or the tokens used for an effect should. * We will store this effect difference in the `InvokeSubgraphCache` where the key is the identifier and value is the effect. For now we only support one effect within a subgraph. * During the tracing part of AOTAutograd, we will then wrap the subgraph to take in and output a token. Before: ``` def forward(self, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', x) return invoke_subgraph def repeated_subgraph(self, x): record_memory = torch.ops.mylib.record_memory.default("forward", "N") add = torch.ops.aten.add(x, x) return add ``` After: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` * Then there is a bunch of logic within `_remove_effect_tokens` to handle removing the effects from the invoke_subgraph subgraph Differential Revision: [D87392741](https://our.internmc.facebook.com/intern/diff/D87392741) Pull Request resolved: #167231 Approved by: https://github.com/anijain2305

…67245) In the [previous PR](https://github.com/pytorch/pytorch/pull/167231/files#diff-e2b74af5d8b538a7d07d18507d27010703742ddad5f819992b55f5abc6d9a502R964-R966) we found that the autograd eager impl of invoke_subgraph calls the subgraph twice. If the subgraph contains effects then effects will be run twice, which is bad. This PR fixes the issue by getting the output metadata from `subgraph`'s `node.meta` if it exists. Differential Revision: [D87392740](https://our.internmc.facebook.com/intern/diff/D87392740) Pull Request resolved: #167245 Approved by: https://github.com/anijain2305 ghstack dependencies: #167231

Updates the implementation of `unlift_tokens` to handle unlifting invoke_subgraph. The context of `unlift_tokens` is currently tokens are threaded as inputs and outputs of the toplevel graph produced by AOTAutograd. However we don't want the inductor traced graph to have any notion of effects/tokens, just that the tokens should introduce some extra dependency behavior. So, we unlift the tokens from the toplevel graph. Instead of placeholder nodes the tokens will come from a `_make_token` call, and instead of outputting the tokens we will sink all tokens into `_sink_tokens`. Similarly, we want the invoke_subgraph subgraph to not have any notion of tokens, so we will also remove the tokens from the inputs of the invoke_subgraph subgraph. However, we still need a way mark the invoke_subgraph call as being effectful at the toplevel module to prevent invoke_subgraph calls from being reordered, so I wrap the invoke_subgraph with an effects. Before: ``` def forward(self, token, x): repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.invoke_subgraph(repeated_subgraph0, 'subgraph_0', token, x) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] return (getitem, getitem_1) def repeated_subgraph(self, token, x): with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) return (getitem, add) ``` After: ``` def forward(self, x): token = torch.ops.prims._make_token.default() repeated_subgraph0 = self.repeated_subgraph0 invoke_subgraph = torch.ops.higher_order.with_effects( token, torch.ops.higher_order.invoke_subgraph, repeated_subgraph0, 'subgraph_0', token, x ) getitem = invoke_subgraph[0] # output token getitem_1 = invoke_subgraph[1] _ = torch.ops.prims._sink_tokens.default([getitem]) return (getitem_1,) def repeated_subgraph(self, x): token = torch.ops.prims._make_token.default() with_effects = torch.ops.higher_order.with_effects(token, torch.ops.mylib.record_memory.default, 'forward', 'N') getitem = with_effects[0] # output token add = torch.ops.aten.add(x, x) _ = torch.ops.prims._sink_tokens.default([getitem]) return (add,) ``` Differential Revision: [D87668981](https://our.internmc.facebook.com/intern/diff/D87668981) Pull Request resolved: #167363 Approved by: https://github.com/fxdawnn ghstack dependencies: #167231, #167245

ghstack-source-id: b97d03a Pull Request resolved: pytorch/pytorch#167231

ghstack-source-id: 251aa71 Pull Request resolved: pytorch/pytorch#167231

Summary: Previously pytorch#167231 modified ep.module() such that it deepcopies ep before applying ep.module() so that ep.module() does not affect the original ep. However we didn't deepcopy the graph signature which caused some issues. This PR addresses it. We also add an additional remove_effect_tokens_pass before returning the new ep so that this ep can be serialized by sigmoid. Differential Revision: D89209276

Previously #167231 modified ep.module() such that it deepcopies ep before applying ep.module() so that ep.module() does not affect the original ep. However we didn't deepcopy the graph signature which caused some issues. Differential Revision: D89209276 Pull Request resolved: #170461 Approved by: https://github.com/ydwu4

Previously pytorch#167231 modified ep.module() such that it deepcopies ep before applying ep.module() so that ep.module() does not affect the original ep. However we didn't deepcopy the graph signature which caused some issues. Differential Revision: D89209276 Pull Request resolved: pytorch#170461 Approved by: https://github.com/ydwu4

[hoo] Invoke subgraph + effect

de4053b

[ghstack-poisoned]

angelayi requested review from avikchaudhuri, tugsbayasgalan, ydwu4, zhxchen17 and zou3519 as code owners November 6, 2025 17:16

This was referenced Nov 4, 2025

[effects] Add register_effectful_op #163284

Closed

[opaque obj] Allow non-effectful scriptobjs #163714

Closed

pytorch-bot bot added the release notes: export label Nov 6, 2025

This was referenced Nov 4, 2025

[opaque obj] torch.compile support #163936

Closed

[opqaue obj] Add attribute support #167230

Closed

facebook-github-bot added the fx label Nov 6, 2025

angelayi added a commit that referenced this pull request Nov 6, 2025

[hoo] Invoke subgraph + effect

1c7498c

ghstack-source-id: 0df0ad4 Pull Request resolved: #167231

Update on "[hoo] Invoke subgraph + effect"

bf3bc66

cc ezyang EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]

angelayi mentioned this pull request Nov 6, 2025

[invoke_subgraph] Don't run the graph twice when autograd enabled #167245

Closed

angelayi marked this pull request as draft November 6, 2025 22:39

Update on "[hoo] Invoke subgraph + effect"

3eb19e6

cc ezyang EikanWang jgong5 wenzhe-nrv [ghstack-poisoned]

angelayi mentioned this pull request Nov 7, 2025

[hoo] Fix unlift of effects with invoke_subgraph #167363

Closed

pytorch-bot bot mentioned this pull request Nov 7, 2025

[hoo] Invoke subgraph + effects Inductor support #167364

Closed

angelayi requested a review from anijain2305 November 7, 2025 21:20

angelayi marked this pull request as ready for review November 7, 2025 21:28

angelayi added 2 commits November 7, 2025 15:42

anijain2305 reviewed Nov 13, 2025

View reviewed changes

anijain2305 approved these changes Nov 13, 2025

View reviewed changes

This was referenced Nov 14, 2025

intermediate, mutation #167877

Closed

test #167878

Closed

test #167879

Closed

pytorchmergebot added Reverted ci-no-td Do not run TD on this PR labels Nov 20, 2025

pytorchmergebot reopened this Nov 20, 2025

angelayi mentioned this pull request Nov 21, 2025

[effect] Remove special handling for profiler op #168389

Closed

pytorchmergebot closed this in 9708e04 Nov 24, 2025

tiendatngcs pushed a commit to tiendatngcs/pytorch-Dec25 that referenced this pull request Dec 10, 2025

[hoo] Invoke subgraph + effect

e183bbf

ghstack-source-id: b97d03a Pull Request resolved: pytorch/pytorch#167231

tiendatngcs pushed a commit to tiendatngcs/pytorch-Dec25 that referenced this pull request Dec 10, 2025

[hoo] Invoke subgraph + effect

390bd61

ghstack-source-id: 251aa71 Pull Request resolved: pytorch/pytorch#167231

angelayi mentioned this pull request Dec 15, 2025

[export] Deepcopy graph signature when unlifting #170461

Closed

github-actions bot deleted the gh/angelayi/132/head branch December 25, 2025 02:20

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[hoo] Invoke subgraph + effect#167231

[hoo] Invoke subgraph + effect#167231
angelayi wants to merge 10 commits intogh/angelayi/132/basefrom
gh/angelayi/132/head

angelayi commented Nov 6, 2025 •

edited

Loading

Uh oh!

pytorch-bot bot commented Nov 6, 2025 •

edited

Loading

Uh oh!

anijain2305 Nov 13, 2025

Uh oh!

angelayi Nov 18, 2025 •

edited

Loading

Uh oh!

pytorchmergebot commented Nov 20, 2025

Uh oh!

pytorchmergebot commented Nov 20, 2025

Uh oh!

angelayi commented Nov 21, 2025

Uh oh!

angelayi commented Nov 21, 2025

Uh oh!

pytorchmergebot commented Nov 24, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

angelayi commented Nov 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Nov 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/167231

Uh oh!

anijain2305 Nov 13, 2025

Choose a reason for hiding this comment

Uh oh!

angelayi Nov 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pytorchmergebot commented Nov 20, 2025

Uh oh!

pytorchmergebot commented Nov 20, 2025

Uh oh!

angelayi commented Nov 21, 2025

Uh oh!

angelayi commented Nov 21, 2025

Uh oh!

pytorchmergebot commented Nov 24, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

angelayi commented Nov 6, 2025 •

edited

Loading

pytorch-bot bot commented Nov 6, 2025 •

edited

Loading

angelayi Nov 18, 2025 •

edited

Loading