[Reasoning] Add thinking budget support

[rishitdholakia13]

…e methods

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.

Purpose

This PR adds thinking budget support for reasoning models. This support has been request as per #15418 and helps build upon changes from #20859 . This PR helps in the following:

• Adds support for models that support speculative and non-speculative decoding
• Avoids the need to use logits processor.
• Uses thinking tokens from reasoningparser
• Allows free thinking, if thinking_bedget is set to -1, then we allow the model think freely without any thinking budget

Test Plan

Test Result

(Optional) Documentation Update

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Summary of Changes

Hello @rishitdholakia13, 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 a 'thinking budget' feature to vLLM, designed to control the token generation length for reasoning models during specific 'thinking' phases. It integrates this functionality across the core scheduling and token generation components, allowing models to adhere to a predefined budget for their internal reasoning steps by leveraging special tokens and dynamically enforcing limits during the sampling process.

Highlights

• Thinking Budget Introduction: Introduced the concept of a 'thinking budget' for reasoning models, allowing control over the number of tokens generated during a dedicated 'thinking' phase. This builds upon existing requests and changes related to reasoning parsers.
• Scheduler Enhancements: Modified the Scheduler to accept and utilize special_token_ids (start/end thinking tokens) and a global thinking_budget. A new utility method get_current_usage was added to track how many thinking tokens a request has consumed. The scheduler now identifies and outputs requests that still have a remaining thinking budget.
• Output Structure Update: The SchedulerOutput class was extended to include requests_with_remaining_budget (a dictionary mapping request IDs to their remaining budget) and end_thinking_token_id, enabling the downstream components to enforce the budget.
• Model Runner Integration: Implemented a new private method _force_thinking in GpuModelRunner. This method actively enforces the thinking budget by truncating sampled tokens for requests that exceed their allocated budget and appending the end_thinking_token_id.
• Configuration Flow: Updated EngineCore to retrieve special_token_ids from the model's reasoning parser and the thinking_budget from the vllm_config, ensuring these parameters are correctly passed down to the Scheduler.

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 is currently in preview and 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 to provide feedback.

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]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @rishitdholakia13.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @rishitdholakia13.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[gemini-code-assist]

Code Review

This PR introduces thinking budget support for reasoning models, allowing for speculative and non-speculative decoding while avoiding the use of logits processors. The changes involve modifications to the scheduler, engine core, and GPU model runner to incorporate thinking tokens and budget management. The code generally looks good, but there are a few areas where improvements can be made to enhance clarity and correctness.