[Bugfix] Fix input processor for InternVL2 model

[Isotr0py]

FILL IN THE PR DESCRIPTION HERE

FIX #7160
FIX #7272

• This PR also aims to make some small refactor to fix some hidden issues. So I marked it as a draft.
• Since most of process args can be obtained from config, this PR removes all fixed args used in InternVL input processor/mapper.

BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE

---

PR Checklist (Click to Expand)

Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.

PR Title and Classification

Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:

• [Bugfix] for bug fixes.
• [CI/Build] for build or continuous integration improvements.
• [Doc] for documentation fixes and improvements.
• [Model] for adding a new model or improving an existing model. Model name should appear in the title.
• [Frontend] For changes on the vLLM frontend (e.g., OpenAI API server, LLM class, etc.)
• [Kernel] for changes affecting CUDA kernels or other compute kernels.
• [Core] for changes in the core vLLM logic (e.g., LLMEngine, AsyncLLMEngine, Scheduler, etc.)
• [Hardware][Vendor] for hardware-specific changes. Vendor name should appear in the prefix (e.g., [Hardware][AMD]).
• [Misc] for PRs that do not fit the above categories. Please use this sparingly.

Note: If the PR spans more than one category, please include all relevant prefixes.

Code Quality

The PR need to meet the following code quality standards:

• We adhere to Google Python style guide and Google C++ style guide.
• Pass all linter checks. Please use format.sh to format your code.
• The code need to be well-documented to ensure future contributors can easily understand the code.
• Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
• Please add documentation to docs/source/ if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.

Notes for Large Changes

Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with rfc-required and might not go through the PR.

What to Expect for the Reviews

The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:

• After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
• After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.
• After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
• Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.

Thank You

Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which consists a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of default ones by unblocking the steps in your fast-check build on Buildkite UI.

Once the PR is approved and ready to go, please make sure to run full CI as it is required to merge (or just use auto-merge).

To run full CI, you can do one of these:

• Comment /ready on the PR
• Add ready label to the PR
• Enable auto-merge.

🚀

[Isotr0py]

/ready

[ywang96]

Thank you for making this PR! @Isotr0py

[ywang96]

Was this the root cause of the original bug?

[Isotr0py]

Yes, because we only append thumbnail image when processed image patches is more than one:

vllm/vllm/model_executor/models/internvl.py

Lines 134 to 137 in 57f560a

	if use_thumbnail and len(processed_images) != 1:
	thumbnail_img = image.resize((image_size, image_size))
	processed_images.append(thumbnail_img)
	return processed_images

[Isotr0py]

Ooops, the root cause should be these lines (L196-L198):

        min_num = hf_config.min_dynamic_patch
        max_num = hf_config.max_dynamic_patch
        num_blocks, _, _ = calculate_num_blocks(width, height, min_num,
                                                max_num, image_size)
        # add thumbnail image if num_blocks > 1
        if hf_config.use_thumbnail and num_blocks > 1:
            num_blocks += 1

The if use_thumbnail: commented above should be OK because we are calculating max image tokens for profiling there, which means len(processed_images) (equal to "max_dynamic_patch" in hf_config) should be larger than 1.

[Isotr0py]

Hmmm, seems that the test is broken. I will check it out tomorrow.

[Isotr0py]

Well, the vision test has passed now.

[DarkLight1337]

I'm going to expand the test suite to cover the problematic model (and also see whether 4B works now).

[DarkLight1337]

It seems like the tests still fail for those models.

[Isotr0py]

Seems that the 26B is running out of VRAM. I have some ideas about testing the problematic 26B model:

1. We can test the InternViT-6B encoder manually, because 26B and larger model only have a different ViT encoder, and their llm backbone have same architecture with small ones.
2. We can try to test 26B with AWQ quantization which will be added in [Model] Add AWQ quantization support for InternVL2 model #7187.

[DarkLight1337]

Seems that the 26B is running out of VRAM. I have some ideas about testing the problematic 26B model:

1. We can test the InternViT-6B encoder manually, because 26B and larger model only have a different ViT encoder, and their llm backbone have same architecture with small ones.
2. We can try to test 26B with AWQ quantization which will add in [Model] Add AWQ quantization support for InternVL2 model #7187.

Is the 4B model fixed by this PR as well? If so, we can just test the 4B version.

[Isotr0py]

No. Instead, the 4B issue is about the Phi3 implementation in the original model repo. Because their Phi3 implementation is out of date and only compatible with transformers==4.37.2.
(You can check this discussion for details: https://huggingface.co/OpenGVLab/InternVL2-4B/discussions/3)

I'm afraid that the only way to fix 4B test is waiting them update their Phi3 implementation to be compatible with transformers >= 4.43.2.

[DarkLight1337]

No. Instead, the 4B issue is about the Phi3 implementation in the original model repo. Because their Phi3 implementation is out of date and only compatible with transformers==4.37.2. (You can check this discussion for details: https://huggingface.co/OpenGVLab/InternVL2-4B/discussions/3)

I'm afraid that the only way to fix 4B test is waiting them update their Phi3 implementation to be compatible with transformers >= 4.43.2.

I see, thanks for the information. I guess we can leave the 26B test for #7187 then.

[Isotr0py]

The internvl model repo uploaded a strange extremely large zip file in their latest commit: https://huggingface.co/OpenGVLab/InternVL2-1B/commit/3b5f67e874a5645a7434b9dbab70a4bc4a3cdf82

Seems that it's an unrelated file for test and significantly slow down the test, I add allow_patterns to exclude it in snapshot_download.

[sfbemerk]

Thanks for your effort! :)
Any ideas when the next version will be released where this fix is included?

[DarkLight1337]

Probably in around 2 weeks. You can build from source if you need it sooner.