[Bugfix] Fix max image size for PaddleOCR-VL

[ywang96]

Purpose

This PR fixes the bug in dummy data generation for the image data that would result in the max number of features/embeddings as well as cleaning up some code since it's always assumed that resizing is turned on.

• On main

Encoder cache will be initialized with a budget of 2048 tokens, and profiled with 10 image items of the maximum feature size.

Max image embedding size is 196.

• This branch

Encoder cache will be initialized with a budget of 3600 tokens, and profiled with 1 image items of the maximum feature size.

Max image embedding size is 3600.

Test Plan

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[gemini-code-assist]

Code Review

This pull request aims to fix a bug in dummy data generation for PaddleOCR-VL to correctly calculate the maximum image embedding size. The changes involve removing a redundant if condition for resizing and updating the logic in get_image_size_with_most_features.

While the intention is correct, I've found a flaw in the new logic for get_image_size_with_most_features. The current implementation is brittle and only works correctly if the maximum number of tokens is a perfect square. I've provided a critical review comment with a more robust implementation to ensure it works for all cases.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[ywang96]

/gemini review

[gemini-code-assist]

Code Review

This pull request fixes a bug in calculating the maximum image size for PaddleOCR-VL models, which is important for memory profiling. The change correctly computes the image dimensions that yield the maximum number of features based on max_pixels. Additionally, it cleans up some redundant code related to image resizing. The core logic of the fix is sound, but I've identified a potential edge case that could lead to a crash.