[Observers] Allow for case when weight shape and block size are not evenly divisble

[brian-dellabetta]

SUMMARY:
Update observer logic for block strategy when weight shape is not divisible by block size

Prerequisite:

• FP8 Block Quantization: Non-Divisible Shape Support compressed-tensors#547

TEST PLAN:

• Quantized checkpoint made with this branch (and above CT branch) runs on vllm main for flashinfer, deepgemm and default kernels -- https://huggingface.co/bdellabe/DeepSeek-V2-Lite-FP8-BLOCK

Run script below with

• VLLM_BLOCKSCALE_FP8_GEMM_FLASHINFER=1 VLLM_USE_DEEP_GEMM=1 for flashinfer
• VLLM_BLOCKSCALE_FP8_GEMM_FLASHINFER=0 VLLM_USE_DEEP_GEMM=1 for deepgemm
• VLLM_BLOCKSCALE_FP8_GEMM_FLASHINFER=0 VLLM_USE_DEEP_GEMM=0 for default

if __name__ == "__main__":
    from vllm import LLM, SamplingParams

    prompts = ["The Swiss Alps are", "Brad Marchand is", "The Toronto Maple Leafs are"]

    # Create a sampling params object for greedy sampling
    sampling_params = SamplingParams(
        temperature=0.80, top_p=0.95, max_tokens=40, min_tokens=10
    )
    llm = LLM(
        "bdellabe/DeepSeek-V2-Lite-FP8-BLOCK",
        max_model_len=4096,
        enforce_eager=True,
    )
    output = llm.generate(prompts, sampling_params)
    for out in output:
        print(out.outputs[0].text)

    print("COMPLETE")

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.

[gemini-code-assist]

Summary of Changes

Hello @brian-dellabetta, 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 enhances the robustness of the observer logic within the llmcompressor library, specifically for block quantization strategies. It introduces a mechanism to gracefully handle scenarios where the dimensions of a weight tensor are not perfectly divisible by the specified block size. By implementing a padding strategy that uses the median values of existing blocks, the system can now process such tensors without error, ensuring broader applicability of the block strategy. Additionally, error handling has been improved by converting assert False statements into explicit ValueError exceptions.

Highlights

• Robust Block Strategy Handling: Implemented padding logic in the _flatten_weight function to correctly handle cases where the weight tensor's dimensions are not evenly divisible by the specified block size, preventing errors in block quantization.
• Median Value Padding: The padding mechanism utilizes the median values of existing blocks to fill the newly added rows and columns, aiming to maintain data integrity and statistical properties.
• Improved Error Handling: Replaced assert False statements with raise ValueError across _flatten_weight, _flatten_activation, and _flatten_attention functions for unknown strategies, providing more explicit and catchable error messages.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

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.

Footnotes

1. 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. ↩

[mergify]

The quality checks have failed. Please run make style and make quality under
the root directory to adddress the lint failures. You will need to install the
dev optional install to get the required linting packages:
https://github.com/vllm-project/llm-compressor/blob/main/CONTRIBUTING.md

[mergify]

The quality checks have failed. Please run make style and make quality under
the root directory to adddress the lint failures. You will need to install the
dev optional install to get the required linting packages:
https://github.com/vllm-project/llm-compressor/blob/main/CONTRIBUTING.md

[gemini-code-assist]

Code Review

This pull request introduces logic to handle cases where weight shapes are not evenly divisible by the block size for block quantization. The core idea of padding the tensor is correct, but the implementation in _flatten_weight has several critical bugs related to padding calculation, tensor creation, and the padding fill logic itself. I've provided a detailed comment with a suggested fix that uses a more standard and robust approach with torch.nn.functional.pad. The other changes to replace assert False with raise ValueError are a good improvement.

[brian-dellabetta]

/gemini review

[gemini-code-assist]

Code Review

This pull request introduces a mechanism to handle cases where weight shapes are not evenly divisible by the block size in block quantization. This is achieved by padding the tensors with the mean of the values in the block being padded, which is a sensible approach to avoid distorting quantization parameters. The changes include a new helper function for padding and updates to the weight flattening logic. The accompanying tests are thorough and cover various scenarios. My main feedback is on the complexity of the new padding function, for which I've suggested a minor refactoring to improve readability.

[brian-dellabetta]

Closing in favor of #2290

[HDCharles]

seems fine