[core][compiled-graphs] Register custom serializers for InputAttributeNodes and pass their type hints to downstream nodes #49236
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
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
kevin85421
changed the title
TorchTensorType does not move tensors to GPU
kevin85421
changed the title
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
kevin85421
commented
Dec 13, 2024
| @@ -1088,6 +1092,9 @@ def _preprocess(self) -> None: | |||
| ): | |||
| downstream_actor_handle = dag_node._get_actor_handle() | |||
|
|
|||
| # Add the type hint of the upstream node to the task. | |||
| task.arg_type_hints.append(upstream_task.dag_node.type_hint) | |||
There was a problem hiding this comment.
Resolve "Reason 1". In the logic below, if the upstream_task.dag_node is an InputAttributeNode, the upstream_task will be set to self.idx_to_task[self.input_task_idx]. This is why I moved this line above.
python/ray/dag/compiled_dag_node.py
Outdated
| input_task.dag_node.type_hint.register_custom_serializer() | ||
|
|
||
| # Register custom serializers for input attribute nodes. | ||
| for input_attr_task_idx in self.input_attr_task_idx_list: |
Signed-off-by: Kai-Hsun Chen <[email protected]>
Signed-off-by: Kai-Hsun Chen <[email protected]>
ruisearch42
approved these changes
Dec 13, 2024
Signed-off-by: Kai-Hsun Chen <[email protected]>
ujjawal-khare
pushed a commit
to ujjawal-khare-27/ray
that referenced
this pull request
Dec 17, 2024
…eNodes and pass their type hints to downstream nodes (ray-project#49236) Signed-off-by: Kai-Hsun Chen <[email protected]> Signed-off-by: ujjawal-khare <[email protected]>
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.
Why are these changes needed?
Issue #47251 states that the tensor will not be moved to the GPU even if users specify
TorchTensorTypeon the input attribute node. The behavior is inconsistent with specifyingTorchTensorTypeon theInputNode.There are two reasons contributing to this inconsistent behavior:
InputAttributeNode's type hint is not passed to the downstream DAG nodes.InputAttributeNode.There are four changes in this PR:
InputNode,InputAttributeNode, and output nodes in a single location.Related issue number
Closes #47251
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.method in Tune, I've added it in
doc/source/tune/api/under thecorresponding
.rstfile.