Skip to content

Changed weight map to tensor and fix the refit bug #3573

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 5 commits into
base: main
Choose a base branch
from

Conversation

cehongwang
Copy link
Collaborator

Description

Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change.

Fixes # (issue)

Type of change

Please delete options that are not relevant and/or add your own.

  • Bug fix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)
  • Breaking change (fix or feature that would cause existing functionality to not work as expected)
  • This change requires a documentation update

Checklist:

  • My code follows the style guidelines of this project (You can use the linters)
  • I have performed a self-review of my own code
  • I have commented my code, particularly in hard-to-understand areas and hacks
  • I have made corresponding changes to the documentation
  • I have added tests to verify my fix or my feature
  • New and existing unit tests pass locally with my changes
  • I have added the relevant labels to my PR in so that relevant reviewers are notified

@github-actions github-actions bot added component: conversion Issues re: Conversion stage component: api [Python] Issues re: Python API component: dynamo Issues relating to the `torch.compile` or `torch._dynamo.export` paths labels Jun 13, 2025
@github-actions github-actions bot requested a review from peri044 June 13, 2025 22:00
# Used for refit
ctx.weight_refit_map[name + " CONSTANT"] = numpy_value.reshape(-1)
ctx.weight_refit_map[name + " CONSTANT"] = torch_value
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Comment the reason of why adding the " Constant"

@cehongwang cehongwang force-pushed the refit-map-type-change branch from d7c6735 to cf064c5 Compare June 13, 2025 22:34
@github-actions github-actions bot added the component: converters Issues re: Specific op converters label Jun 14, 2025
@@ -321,7 +321,15 @@ def cast_int_or_float_to_bool(


def to_trt_weights(
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we can streamline some arguments like why do we need target, name, layer name and weight name? can we derive some of these from others?

@cehongwang cehongwang force-pushed the refit-map-type-change branch from 41d1248 to 2520a68 Compare June 16, 2025 21:08
cpu_weights_reference_holder: dict[str, Union[torch.Tensor]] = field(
default_factory=dict
)

def record_weight(self, name: str, weight: torch.Tensor) -> None:
self.weight_refit_map[name] = weight
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

add a docstring explaining why we are doing this especially the comment related to self.cpu_weights_reference_holder[name + " CPU_REFERENCE"] = weight

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The name " CPU_REFERENCE" is a bit random. Any name could work because all we need is to hold it on CPU. Moreover, since we have refit map, this is actually a bit redundant.

Copy link
Collaborator

@narendasan narendasan Jun 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do we need a suffix here? Its just holding references, it should not be inspected later. I dont think it should even be a dictionary

supported_weight_types = ["KERNEL", "BIAS", "CONSTANT"]
assert (
layer_type_name in supported_layer_types
), f"Unsupported layer type: {layer_type_name}. Please add the layer type to this function to enable refitting."
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please add the layer type to this function to enable refitting. - what does this mean ? How do we add this ?

ctx: ConversionContext,
value: torch.Tensor,
name: str,
layer_type_name: str,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should use those literal type annotations

weight_type_name in supported_weight_types
), f"Encountered unsupported weight type: {weight_type_name}. Supported types are: {supported_weight_types}. Manually calling to_trt_weights with a custom weight type is not intended for general use."

if weight_type_name == "CONSTANT" and layer_type_name == "CONSTANT":
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What is the difference between a weight type and a layer type?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
cla signed component: api [Python] Issues re: Python API component: conversion Issues re: Conversion stage component: converters Issues re: Specific op converters component: dynamo Issues relating to the `torch.compile` or `torch._dynamo.export` paths
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants