Avoid FX Constant Folding in rw_sharding #2777
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
facebook-github-bot
added
the
CLA Signed
cp2923
changed the title
|
This pull request was exported from Phabricator. Differential Revision: D70577218 |
|
This pull request was exported from Phabricator. Differential Revision: D70577218 |
cp2923
changed the title
cp2923
pushed a commit
to cp2923/torchrec
that referenced
this pull request
Mar 5, 2025
Summary: Tensor_constants nodes are known to have issues with delta update, that the predictor cannot find the tensors in the delta weights. To enable the delta publish works for blue_reels_vdd/v5, we need to ensure the model graph does not containing tensor_constants nodes Breakdown to two diffs, this one contains the file change requires OSS pull request | Constant tensor name | fx paste | code file | code | | -- | | merge._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?color=0&handle=GLJIlhayl6fBeToDACvUvTwjbsMkbr0LAAAz | model_hstu_standalone_cint.py | torch.tensor(self.tasks_with_uid_emb_indices).to(r_over.device) | | merge.main_module.module.sim_module.module_impl._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?handle=GG71xRydBrkmAncFAPQ_36FF6X8XbsIXAAAB | sim.py | self.action_weights[None, : ] | | Input_dist._tensor_constant0, ..1, ..2, etc | https://www.internalfb.com/intern/everpaste/?color=0&handle=GG7hnhaueOUr4MIFAAViKnOmlYozbr0LAAAz | rw_sharding.py | torch.tensor(block_sizes, ...) | Differential Revision: D70577218
|
This pull request was exported from Phabricator. Differential Revision: D70577218 |
cp2923
pushed a commit
to cp2923/torchrec
that referenced
this pull request
Mar 5, 2025
Summary: Pull Request resolved: pytorch#2777 Tensor_constants nodes are known to have issues with delta update, that the predictor cannot find the tensors in the delta weights. To enable the delta publish works for blue_reels_vdd/v5, we need to ensure the model graph does not containing tensor_constants nodes Breakdown to two diffs, this one contains the file change requires OSS pull request | Constant tensor name | fx paste | code file | code | | -- | | merge._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?color=0&handle=GLJIlhayl6fBeToDACvUvTwjbsMkbr0LAAAz | model_hstu_standalone_cint.py | torch.tensor(self.tasks_with_uid_emb_indices).to(r_over.device) | | merge.main_module.module.sim_module.module_impl._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?handle=GG71xRydBrkmAncFAPQ_36FF6X8XbsIXAAAB | sim.py | self.action_weights[None, : ] | | Input_dist._tensor_constant0, ..1, ..2, etc | https://www.internalfb.com/intern/everpaste/?color=0&handle=GG7hnhaueOUr4MIFAAViKnOmlYozbr0LAAAz | rw_sharding.py | torch.tensor(block_sizes, ...) | Differential Revision: D70577218
cp2923
pushed a commit
to cp2923/torchrec
that referenced
this pull request
Mar 6, 2025
Summary: Tensor_constants nodes are known to have issues with delta update, that the predictor cannot find the tensors in the delta weights. To enable the delta publish works for blue_reels_vdd/v5, we need to ensure the model graph does not containing tensor_constants nodes Breakdown to two diffs, this one contains the file change requires OSS pull request | Constant tensor name | fx paste | code file | code | | -- | | merge._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?color=0&handle=GLJIlhayl6fBeToDACvUvTwjbsMkbr0LAAAz | model_hstu_standalone_cint.py | torch.tensor(self.tasks_with_uid_emb_indices).to(r_over.device) | | merge.main_module.module.sim_module.module_impl._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?handle=GG71xRydBrkmAncFAPQ_36FF6X8XbsIXAAAB | sim.py | self.action_weights[None, : ] | | Input_dist._tensor_constant0, ..1, ..2, etc | https://www.internalfb.com/intern/everpaste/?color=0&handle=GG7hnhaueOUr4MIFAAViKnOmlYozbr0LAAAz | rw_sharding.py | torch.tensor(block_sizes, ...) | Differential Revision: D70577218
|
This pull request was exported from Phabricator. Differential Revision: D70577218 |
cp2923
pushed a commit
to cp2923/torchrec
that referenced
this pull request
Mar 6, 2025
Summary: Pull Request resolved: pytorch#2777 Tensor_constants nodes are known to have issues with delta update, that the predictor cannot find the tensors in the delta weights. To enable the delta publish works for blue_reels_vdd/v5, we need to ensure the model graph does not containing tensor_constants nodes Breakdown to two diffs, this one contains the file change requires OSS pull request | Constant tensor name | fx paste | code file | code | | -- | | merge._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?color=0&handle=GLJIlhayl6fBeToDACvUvTwjbsMkbr0LAAAz | model_hstu_standalone_cint.py | torch.tensor(self.tasks_with_uid_emb_indices).to(r_over.device) | | merge.main_module.module.sim_module.module_impl._tensor_constant0 | https://www.internalfb.com/intern/everpaste/?handle=GG71xRydBrkmAncFAPQ_36FF6X8XbsIXAAAB | sim.py | self.action_weights[None, : ] | | Input_dist._tensor_constant0, ..1, ..2, etc | https://www.internalfb.com/intern/everpaste/?color=0&handle=GG7hnhaueOUr4MIFAAViKnOmlYozbr0LAAAz | rw_sharding.py | torch.tensor(block_sizes, ...) | Differential Revision: D70577218
Summary:
Tensor_constants nodes are known to have issues with delta update, that the predictor cannot find the tensors in the delta weights.
'tensor_cache' in this diff introduces tenor constant in fx.
Solution:
The solution is to replace 'tensor_cache' through 'register_buffer' + move device and data_type at first forward (so as to keep performance parity).
Note this solution breaks a unit test
```
assertFalse(
hasattr(
local.ro_ec,
"_root_mc_embedding_collection",
)
)
```
but it still won't introduce big table like tbes.
Reviewed By: jingsh
Differential Revision: D70577218
cp2923
force-pushed
the
export-D70577218
branch
from
19e17fd to
2f83768
Compare
|
This pull request was exported from Phabricator. Differential Revision: D70577218 |
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:
Tensor_constants nodes are known to have issues with delta update, we need to ensure the model graph does not containing tensor_constants nodes
| Constant tensor name | code file | code |
| -- |
| Input_dist._tensor_constant0, ..1, ..2, etc | rw_sharding.py | torch.tensor(block_sizes, ...) |
Differential Revision: D70577218