feat(transformers): add hgnet_v2 (v4.54.1) #1395

alien-0119 · 2025-10-27T08:39:14Z

What does this PR do?

Adds # (feature)
Add hgnet_v2 model and fast ut.

Usage Example:

import mindspore as ms
import requests
from mindone.transformers import HGNetV2ForImageClassification
from transformers import AutoImageProcessor
from PIL import Image


url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)

model = HGNetV2ForImageClassification.from_pretrained("ustc-community/hgnet-v2")
processor = AutoImageProcessor.from_pretrained("ustc-community/hgnet-v2")

inputs = processor(images=image, return_tensors="np")
inputs = {k: ms.tensor(v).to(dtype) for k, v in inputs.items()}

logits = model(**inputs).logits
predicted_class_id = logits.argmax(dim=-1).item()

class_labels = model.config.id2label
predicted_class_label = class_labels[predicted_class_id]
print(f"The predicted class label is: {predicted_class_label}")
# LABEL_0

Performance:
Experiments were tested on Ascend Atlas 800T A2 machines with mindspore 2.7.0 pynative mode.

mindspore.nn.MaxPool2d does not support bf16 inputs

model	precision	weight load(s)	s/step
ustc-community/hgnet-v2	fp32	21.629	0.026
ustc-community/hgnet-v2	fp16	23.190	0.024

Before submitting

This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
Did you read the contributor guideline?
Did you make sure to update the documentation with your changes? E.g. record bug fixes or new features in What's New. Here are the
documentation guidelines
Did you build and run the code without any errors?
Did you report the running environment (NPU type/MS version) and performance in the doc? (better record it for data loading, model inference, or training tasks)
Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@xxx

gemini-code-assist · 2025-10-27T08:39:32Z

Summary of Changes

Hello @alien-0119, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces the HGNetV2 model to the mindone/transformers framework, expanding the library's capabilities in computer vision. The changes encompass the full implementation of the HGNetV2 architecture, its pre-trained model, and a specific variant for image classification. This integration ensures that users can leverage HGNetV2 through the standard transformers API, with robust testing to validate its performance and compatibility within the MindSpore ecosystem.

Highlights

New Model Integration: The HGNetV2 model, including its backbone and image classification capabilities, has been added to the mindone/transformers library.
Auto-Configuration Support: The new HGNetV2 model is fully integrated into the auto-configuration and auto-modeling systems, allowing for easy instantiation and usage within the framework.
Comprehensive Testing: Dedicated tests have been introduced for the HGNetV2 model to ensure functional correctness and precision consistency between PyTorch and MindSpore implementations.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request adds the HGNet-V2 model, including its backbone and image classification head, along with corresponding tests. The implementation is well-structured and integrates with the existing auto-model framework. My review includes a few suggestions for improvement: renaming a parameter that shadows a Python built-in, refactoring a part of the classification head to use more idiomatic MindSpore code, correcting a typo in a test filename, and, most importantly, extending the tests to cover graph mode, which is crucial for ensuring model correctness in MindSpore.

gemini-code-assist · 2025-10-27T08:42:19Z

tests/transformers_tests/models/hgnet_v2/test_modeing_hgnet_v2.py

+
+# ms.nn.MaxPool2d does not support bf16 inputs
+DTYPE_AND_THRESHOLDS = {"fp32": 6e-4, "fp16": 5e-3}
+MODES = [1]


The tests are currently only running in PyNative mode (ms.PYNATIVE_MODE which is 1). It's important to also test in Graph mode (ms.GRAPH_MODE which is 0) to ensure the model can be correctly compiled and executed. Please consider adding it to the MODES list.

Suggested change

MODES = [1]

MODES = [0, 1]

gemini-code-assist · 2025-10-27T08:42:20Z

mindone/transformers/models/hgnet_v2/modeling_hgnet_v2.py

+        else:
+            self.lab = mint.nn.Identity()
+
+    def construct(self, input: Tensor) -> Tensor:


The parameter input shadows a built-in Python function. It's a good practice to avoid this. I suggest renaming it to hidden_state for clarity and consistency with the variable's usage within the method.

Suggested change

def construct(self, input: Tensor) -> Tensor:

def construct(self, hidden_state: Tensor) -> Tensor:

gemini-code-assist · 2025-10-27T08:42:20Z

mindone/transformers/models/hgnet_v2/modeling_hgnet_v2.py

+        )
+
+        # classification head
+        self.classifier = nn.CellList([self.avg_pool, self.flatten])


Using nn.CellList with a for loop in the construct method for a simple sequence of layers is less idiomatic than using nn.SequentialCell. I recommend refactoring this to nn.SequentialCell.

self.classifier = nn.SequentialCell(self.avg_pool, self.flatten)

Then, in the construct method (lines 447-449), you can replace the loop over self.classifier with a single call:

pooled_output = self.classifier(last_hidden_state) logits = self.fc(pooled_output)

This will make the code cleaner and more aligned with common MindSpore practices.

gemini-code-assist · 2025-10-27T08:42:20Z

tests/transformers_tests/models/hgnet_v2/test_modeing_hgnet_v2.py

@@ -0,0 +1,223 @@
+"""Adapted from https://github.com/huggingface/transformers/tree/main/tests/models/hgnet_v2/test_modeling_hgnet_v2.py."""


There seems to be a typo in the filename. It should probably be test_modeling_hgnet_v2.py instead of test_modeing_hgnet_v2.py.

alien-0119 requested a review from vigo999 as a code owner October 27, 2025 08:39

gemini-code-assist bot reviewed Oct 27, 2025

View reviewed changes

alien-0119 force-pushed the hgnet_v2_master branch from fb1af77 to 97ff01f Compare October 27, 2025 08:48

add hgnet_v2

a9bc4f4

alien-0119 force-pushed the hgnet_v2_master branch from 97ff01f to a9bc4f4 Compare October 28, 2025 03:45

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

feat(transformers): add hgnet_v2 (v4.54.1) #1395

feat(transformers): add hgnet_v2 (v4.54.1) #1395

Uh oh!

alien-0119 commented Oct 27, 2025 •

edited

Loading

Uh oh!

gemini-code-assist bot commented Oct 27, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Oct 27, 2025

Uh oh!

gemini-code-assist bot Oct 27, 2025

Uh oh!

gemini-code-assist bot Oct 27, 2025

Uh oh!

gemini-code-assist bot Oct 27, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

	def construct(self, input: Tensor) -> Tensor:
	def construct(self, hidden_state: Tensor) -> Tensor:

		@@ -0,0 +1,223 @@
		"""Adapted from https://github.com/huggingface/transformers/tree/main/tests/models/hgnet_v2/test_modeling_hgnet_v2.py."""

Uh oh!

feat(transformers): add hgnet_v2 (v4.54.1) #1395

Are you sure you want to change the base?

feat(transformers): add hgnet_v2 (v4.54.1) #1395

Uh oh!

Conversation

alien-0119 commented Oct 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Before submitting

Who can review?

Uh oh!

gemini-code-assist bot commented Oct 27, 2025

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Oct 27, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Oct 27, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Oct 27, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Oct 27, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

alien-0119 commented Oct 27, 2025 •

edited

Loading