Replace "joint lists" with a list of objects in ArgInfo #2742
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
|
This pull request was exported from Phabricator. Differential Revision: D69461227 |
che-sh
force-pushed
the
export-D69461227
branch
from
a0f0277 to
3ff0b8a
Compare
|
This pull request was exported from Phabricator. Differential Revision: D69461227 |
che-sh
force-pushed
the
export-D69461227
branch
from
3ff0b8a to
301bbe2
Compare
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 13, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
|
This pull request was exported from Phabricator. Differential Revision: D69461227 |
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 13, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 13, 2025
…Info (pytorch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: * almost all code in `_build_args_kwargs` deals with the fields of ArgInfoStep, and remaining part handles looping over `ArgInfo.steps` - so this change just colocates "behavior" (`_build_args_kwargs` logic) with data it belongs to. * introduces helper functions/factory methods for various types of ArgInfoStep * encapsulates the logic of handling a `List[ArgInfo]` into a `CallArgs` class (+changes a bit - explicitly separating args nad kwargs, vs. having them differ by empty/present `ArgInfo.name` field) Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep (**you are here**) 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461226
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 13, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 13, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 13, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 14, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 18, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 18, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
force-pushed
the
export-D69461227
branch
from
301bbe2 to
28fd5d8
Compare
|
This pull request was exported from Phabricator. Differential Revision: D69461227 |
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 19, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
force-pushed
the
export-D69461227
branch
from
28fd5d8 to
381207d
Compare
|
This pull request was exported from Phabricator. Differential Revision: D69461227 |
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 19, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 19, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 20, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 20, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
force-pushed
the
export-D69461227
branch
from
381207d to
334fea2
Compare
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 20, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
|
This pull request was exported from Phabricator. Differential Revision: D69461227 |
…stproc modules (pytorch#2733) Summary: Postproc modules with collection inputs (list or dict) with non-static (derived from input or other postproc) elements were not properly rewritten - input elements remained fx.Nodes even during the actual model forward (i.e. outside rewrite, during pipeline execution) To illustrate: ``` def forward(model_input: ...) -> ...: modified_input = model_input.float_features + 1 sharded_module_input = self.postproc(model_input, modified_input) # works sharded_module_input = self.postproc(model_input, [123]) # works sharded_module_input = self.postproc(model_input, [torch.ones_like(modified_input)]) # fails sharded_module_input = self.postproc(model_input, [modified_input]) # fails sharded_module_input = self.postproc(model_input, { 'a': 123 }) # works sharded_module_input = self.postproc(model_input, { 'a': torch.ones_like(modified_input) }) # fails sharded_module_input = self.postproc(model_input, { 'a': modified_input }) # fails return self.ebc(sharded_module_input) ``` Reviewed By: sarckk Differential Revision: D69292525
Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: _get_node_args and related functions pass around lot of "context" (train_pipeline_context, streams, etc.) that rarely or never changes + some "state" (model, pipelined_preprocs) that is accumulated during the run. Refactoring `_get_node_args` (and friends) into a class allows initializing/passing those into class constructor, and simplifies the call signatures a lot Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class (**you are here**) 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69438143
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 20, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
che-sh
force-pushed
the
export-D69461227
branch
from
334fea2 to
558c546
Compare
|
This pull request was exported from Phabricator. Differential Revision: D69461227 |
che-sh
added a commit
to che-sh/torchrec
that referenced
this pull request
Feb 20, 2025
…ch#2742) Summary: Torchrec rewriting logic got a bit hairy over the years, this sequence of changes aims to refactor the rewrite logic to be less convoluted and more maintainable in the future. This change: ArgInfo uses a "synchronized lists" pattern, having 4 attributes, each being a list, semantically representing different fields of a data structure (i.e. input_attrs[0], is_getitems[0], ... all relate to a single transformation on the input; all lists must have same number of elements). This diff refactors them into an actual list of a (new) `ArgInfoStep` class instances that encapsulate the related fields. Internal Diff stack navigation: 1. D69292525 and below - before refactoring 2. D69438143 - Refactor get_node_args and friends into a class 3. D69461227 - refactor "joint lists" in ArgInfo into a list of ArgInfoStep (**you are here**) 4. D69461226 - refactor `_build_args_kwargs` into instance methods on ArgInfo and ArgInfoStep 5. D69461228 - split monolithic `ArgInfoStep` into a class hierarchy Differential Revision: D69461227
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.
Differential Revision: D69461227