Clean up attention backend selection code & Other minor rename

[merrymercy]

• move the default attention backend selection logic from model_runner.py to server_args.py, so we can put all attention backend related things into a single place
• rename spec_num_total_accepted_tokens -> spec_num_accepted_tokens , cum_spec_accept_length -> spec_total_num_accepted_tokens

[gemini-code-assist]

Summary of Changes

Hello @merrymercy, 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 focuses on improving the organization and clarity of the codebase. It centralizes the complex logic for attention backend selection within the server_args module, making it easier to manage and understand. Additionally, it renames key metrics used in speculative decoding to enhance their descriptive accuracy, contributing to better maintainability and readability of the performance monitoring system.

Highlights

• Refactoring Attention Backend Selection: The logic for automatically selecting the attention backend has been moved from model_runner.py to server_args.py. This centralizes all attention backend related configurations and decisions into a single, more appropriate location.
• Renaming Speculative Decoding Metrics: Several internal variables related to speculative decoding metrics have been renamed for improved clarity and consistency across the codebase. Specifically, spec_num_total_accepted_tokens is now spec_num_accepted_tokens, spec_num_total_forward_ct is spec_num_forward_ct, cum_spec_accept_length is spec_total_num_accepted_tokens, and cum_spec_accept_count is spec_total_num_forward_ct.

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. ↩

[gemini-code-assist]

Code Review

This pull request refactors the attention backend selection logic by moving it from model_runner.py to server_args.py for better organization, and renames several speculative decoding metric variables for clarity. While the intention is good, the refactoring introduces a critical issue where device-dependent logic is executed before the correct device context is established for worker processes. This can lead to incorrect attention backend selection. Additionally, there's a functional change in a memory-related parameter that might be unintentional.