[FSDP2] Allowed List[nn.Module] as arg
#127786
Closed
Conversation
This file contains 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
[ghstack-poisoned]
This was referenced Jun 3, 2024
awgu
commented
Jun 3, 2024
Sequence[nn.Module] as argList[nn.Module] as arg
cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
awgu
commented
Jun 3, 2024
| ] | ||
| post_handles = [ | ||
| module.register_forward_hook( | ||
| get_wrapped_post_hook(module), prepend=False, always_call=True |
There was a problem hiding this comment.
cc: @yf225 Do you know if torch.compile can handle always_call=True today?
There was a problem hiding this comment.
Previous context is in #127786 (comment)
There was a problem hiding this comment.
I believe it should work with torch.compile. Wonder is this path triggered for typical FSDP2 use cases currently? If
There was a problem hiding this comment.
Ah, this code path is new, so none of the existing compile tests should exercise it. I can do a quick test to see if trying to use this code path with your compile tests works or not.
cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
This was referenced Jun 6, 2024
cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
awgu
added a commit
to awgu/pytorch
that referenced
this pull request
Jun 21, 2024
ghstack-source-id: 55344a83366a1d4eef8b22fd7eb0166696ba1653 Pull Request resolved: pytorch#127786
cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
This PR allows `fully_shard`'s first argument to be `List[nn.Module]` instead of strictly `nn.Module`. This allows more flexible grouping of modules/parameters for communication, which can lead to memory savings and/or more efficient communication. **Approach** At a high level, we can think of a model as a tree of modules. Previously, we could only select specific module nodes in this tree as representing one FSDP parameter group. With this PR, we can select a group of module nodes, effectively becoming a single super node. To implement the runtime schedule, we define new forward hooks that run based on the following semantics: - If a module is the first to run the pre-hook, actually run the given pre-hook. Otherwise, the pre-hook is no-op. - If a module is the last to run the post-hook, actually run the given post-hook. Otherwise, the post-hook is a no-op. - First and last are determined by scoreboarding against a set of the modules. - This set must get cleared at the end of backward in the case that >=1 module in the list is never used, in which case we still want the forward hooks to run in the next forward after this backward. Beyond these new forward hooks, everything else is some simple generalization from `Module` to `List[Module]` or `Tuple[Module, ...]`. **Examples** This PR enables wrapping Llama models more efficiently by grouping the final norm and output linear together: pytorch/torchtitan#382. If at least one of the modules in the list does not run forward before backward, then there will be a warning message like: ``` 1 of the 2 modules passed to fully_shard did not run forward before backward, which is error-prone since FSDP post-forward/pre-backward logic will not run for these modules. We recommend passing only modules that run forward together. Modules that did not run forward: [FSDPLinear(in_features=1, out_features=1, bias=True)] ``` cc mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse H-Huang kwen2501 penguinwu fegin XilunWu wanchaol fduwjj wz337 tianyu-l wconstab yf225 chauhang d4l3k [ghstack-poisoned]
awgu
added a commit
that referenced
this pull request
Jul 15, 2024
ghstack-source-id: 95e1b52892cd6218ecc6628c5898bd788543bb32 Pull Request resolved: #127786
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
|
@pytorchmergebot revert -c ghfirst -m "bottom pr from the stack is failing on internal error" |
|
@pytorchbot successfully started a revert job. Check the current status here. |
|
@awgu your PR has been successfully reverted. |
pytorchmergebot
added a commit
that referenced
this pull request
Jul 16, 2024
This reverts commit d3ab8ce. Reverted #127786 on behalf of https://github.com/atalman due to bottom pr from the stack is failing on internal error ([comment](#127786 (comment)))
awgu
added a commit
that referenced
this pull request
Jul 17, 2024
This reverts commit d3ab8ce. Reverted #127786 on behalf of https://github.com/atalman due to bottom pr from the stack is failing on internal error ([comment](#127786 (comment))) [ghstack-poisoned]
This PR allows `fully_shard`'s first argument to be `List[nn.Module]` instead of strictly `nn.Module`. This allows more flexible grouping of modules/parameters for communication, which can lead to memory savings and/or more efficient communication. **Approach** At a high level, we can think of a model as a tree of modules. Previously, we could only select specific module nodes in this tree as representing one FSDP parameter group. With this PR, we can select a group of module nodes, effectively becoming a single super node. To implement the runtime schedule, we define new forward hooks that run based on the following semantics: - If a module is the first to run the pre-hook, actually run the given pre-hook. Otherwise, the pre-hook is no-op. - If a module is the last to run the post-hook, actually run the given post-hook. Otherwise, the post-hook is a no-op. - First and last are determined by scoreboarding against a set of the modules. - This set must get cleared at the end of backward in the case that >=1 module in the list is never used, in which case we still want the forward hooks to run in the next forward after this backward. Beyond these new forward hooks, everything else is some simple generalization from `Module` to `List[Module]` or `Tuple[Module, ...]`. **Examples** This PR enables wrapping Llama models more efficiently by grouping the final norm and output linear together: pytorch/torchtitan#382. If at least one of the modules in the list does not run forward before backward, then there will be a warning message like: ``` 1 of the 2 modules passed to fully_shard did not run forward before backward, which is error-prone since FSDP post-forward/pre-backward logic will not run for these modules. We recommend passing only modules that run forward together. Modules that did not run forward: [FSDPLinear(in_features=1, out_features=1, bias=True)] ``` cc XilunWu H-Huang kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o mrshenli pritamdamania87 zhaojuanmao satgera gqchen aazzolini osalpekar jiayisuse penguinwu tianyu-l yf225 chauhang [ghstack-poisoned]
awgu
added a commit
that referenced
this pull request
Jul 17, 2024
This PR allows `fully_shard`'s first argument to be `List[nn.Module]` instead of strictly `nn.Module`. This allows more flexible grouping of modules/parameters for communication, which can lead to memory savings and/or more efficient communication. **Approach** At a high level, we can think of a model as a tree of modules. Previously, we could only select specific module nodes in this tree as representing one FSDP parameter group. With this PR, we can select a group of module nodes, effectively becoming a single super node. To implement the runtime schedule, we define new forward hooks that run based on the following semantics: - If a module is the first to run the pre-hook, actually run the given pre-hook. Otherwise, the pre-hook is no-op. - If a module is the last to run the post-hook, actually run the given post-hook. Otherwise, the post-hook is a no-op. - First and last are determined by scoreboarding against a set of the modules. - This set must get cleared at the end of backward in the case that >=1 module in the list is never used, in which case we still want the forward hooks to run in the next forward after this backward. Beyond these new forward hooks, everything else is some simple generalization from `Module` to `List[Module]` or `Tuple[Module, ...]`. **Examples** This PR enables wrapping Llama models more efficiently by grouping the final norm and output linear together: pytorch/torchtitan#382. If at least one of the modules in the list does not run forward before backward, then there will be a warning message like: ``` 1 of the 2 modules passed to fully_shard did not run forward before backward, which is error-prone since FSDP post-forward/pre-backward logic will not run for these modules. We recommend passing only modules that run forward together. Modules that did not run forward: [FSDPLinear(in_features=1, out_features=1, bias=True)] ``` Pull Request resolved: #127786 Approved by: https://github.com/yf225, https://github.com/weifengpy ghstack dependencies: #127773 ghstack-source-id: c70870546bed6108b163062b88415cbba3d37925
|
Closing in favor of #130949 |
mlazos
pushed a commit
that referenced
this pull request
Jul 18, 2024
This reverts commit d3ab8ce. Reverted #127786 on behalf of https://github.com/atalman due to bottom pr from the stack is failing on internal error ([comment](#127786 (comment)))
bigfootjon
pushed a commit
that referenced
this pull request
Jul 24, 2024
This reverts commit d3ab8ce. Reverted #127786 on behalf of https://github.com/atalman due to bottom pr from the stack is failing on internal error ([comment](#127786 (comment))) (cherry picked from commit 4f40a70)
xuhancn
pushed a commit
to xuhancn/pytorch
that referenced
this pull request
Jul 25, 2024
This PR allows `fully_shard`'s first argument to be `List[nn.Module]` instead of strictly `nn.Module`. This allows more flexible grouping of modules/parameters for communication, which can lead to memory savings and/or more efficient communication. **Approach** At a high level, we can think of a model as a tree of modules. Previously, we could only select specific module nodes in this tree as representing one FSDP parameter group. With this PR, we can select a group of module nodes, effectively becoming a single super node. To implement the runtime schedule, we define new forward hooks that run based on the following semantics: - If a module is the first to run the pre-hook, actually run the given pre-hook. Otherwise, the pre-hook is no-op. - If a module is the last to run the post-hook, actually run the given post-hook. Otherwise, the post-hook is a no-op. - First and last are determined by scoreboarding against a set of the modules. - This set must get cleared at the end of backward in the case that >=1 module in the list is never used, in which case we still want the forward hooks to run in the next forward after this backward. Beyond these new forward hooks, everything else is some simple generalization from `Module` to `List[Module]` or `Tuple[Module, ...]`. **Examples** This PR enables wrapping Llama models more efficiently by grouping the final norm and output linear together: pytorch/torchtitan#382. If at least one of the modules in the list does not run forward before backward, then there will be a warning message like: ``` 1 of the 2 modules passed to fully_shard did not run forward before backward, which is error-prone since FSDP post-forward/pre-backward logic will not run for these modules. We recommend passing only modules that run forward together. Modules that did not run forward: [FSDPLinear(in_features=1, out_features=1, bias=True)] ``` Pull Request resolved: pytorch#127786 Approved by: https://github.com/yf225, https://github.com/weifengpy ghstack dependencies: pytorch#127773
xuhancn
pushed a commit
to xuhancn/pytorch
that referenced
this pull request
Jul 25, 2024
This reverts commit d3ab8ce. Reverted pytorch#127786 on behalf of https://github.com/atalman due to bottom pr from the stack is failing on internal error ([comment](pytorch#127786 (comment)))
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.
Stack from ghstack (oldest at bottom):
List[nn.Module]as arg #127786contractto allowSequence[nn.Module](#127773) #130947This PR allows
fully_shard's first argument to beList[nn.Module]instead of strictlynn.Module. This allows more flexible grouping of modules/parameters for communication, which can lead to memory savings and/or more efficient communication.Approach
At a high level, we can think of a model as a tree of modules. Previously, we could only select specific module nodes in this tree as representing one FSDP parameter group. With this PR, we can select a group of module nodes, effectively becoming a single super node.
To implement the runtime schedule, we define new forward hooks that run based on the following semantics:
Beyond these new forward hooks, everything else is some simple generalization from
ModuletoList[Module]orTuple[Module, ...].Examples
This PR enables wrapping Llama models more efficiently by grouping the final norm and output linear together: pytorch/torchtitan#382.
If at least one of the modules in the list does not run forward before backward, then there will be a warning message like:
cc @XilunWu @H-Huang @kwen2501 @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @c-p-i-o @mrshenli @pritamdamania87 @zhaojuanmao @satgera @gqchen @aazzolini @osalpekar @jiayisuse @penguinwu @tianyu-l @yf225 @chauhang