[Bugfix] Fix qwen2.5-vl-without-padding

[Potabk]

What this PR does / why we need it?

Correct AscendQwen2_5_VLForConditionalGeneration_Without_Padding override methods

Does this PR introduce any user-facing change?

How was this patch tested?

• vLLM version: v0.10.1.1
• vLLM main: vllm-project/vllm@42dc59d

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request corrects the class structure for AscendQwen2_5_VLForConditionalGeneration_Without_Padding by properly including the _process_image_input and _process_video_input methods within the class, fixing what appears to be an indentation issue. It also adds tests to verify that these methods are correctly overridden from the parent class.

The changes are correct and address the bug. I have one suggestion to improve the efficiency of the __init__ method in AscendQwen2_5_VLForConditionalGeneration_Without_Padding to avoid unnecessary object instantiation during model loading.

[gemini-code-assist]

The current __init__ method calls super().__init__, which instantiates the parent's Qwen2_5_VisionTransformer. This object is then immediately discarded and replaced by an AscendQwen2_5_VisionTransformer_Without_Padding. This is inefficient and can impact model loading performance and memory usage.

A better approach is to bypass the parent's __init__ and replicate its essential logic. This involves calling nn.Module.__init__() directly and then initializing the language_model attribute, as done for self.visual. This avoids the unnecessary object creation.

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        # Call `nn.Module.__init__` directly to avoid instantiating the base
        # vision model, which is immediately replaced. This improves efficiency.
        nn.Module.__init__(self)

        config: Qwen2_5_VLConfig = vllm_config.model_config.hf_config
        quant_config = vllm_config.quant_config
        self.visual = AscendQwen2_5_VisionTransformer_Without_Padding(
            vision_config=config.vision_config,
            norm_eps=getattr(config, "rms_norm_eps", 1e-6),
            quant_config=self._maybe_ignore_quant_config(quant_config),
            prefix=maybe_prefix(prefix, "visual"),
        )

        # Replicate language_model creation from the parent class.
        # A local import is used to avoid modifying file-level imports.
        from vllm.model_executor.models.qwen2 import Qwen2ForCausalLM
        self.language_model = Qwen2ForCausalLM(config,
                                               quant_config=quant_config,
                                               prefix=prefix)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.42%. Comparing base (6c97336) to head (0a145a6).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2623      +/-   ##
==========================================
+ Coverage   72.35%   72.42%   +0.06%     
==========================================
  Files         146      146              
  Lines       21622    21608      -14     
==========================================
+ Hits        15645    15649       +4     
+ Misses       5977     5959      -18     

Flag	Coverage Δ
unittests	72.42% <100.00%> (+0.06%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.