Resize rope embeddings for long sequence training

[mmathew23]

Fix for #3036 #3216

Llama Notebook Test:
https://colab.research.google.com/drive/1N76XCQ1WsIMjQSbZs8HgTqjCbk64C1-T?usp=sharing

Shows that a sequence length that used to fail now works with the fix branch.

[gemini-code-assist]

Summary of Changes

Hello @mmathew23, 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 critical fix for Llama models, enabling them to handle and train with significantly longer sequence lengths. By dynamically extending the Rotary Position Embeddings (RoPE) within the LlamaModel_fast_forward method, the change resolves previous limitations that caused failures when processing sequences exceeding the initial embedding capacity. This enhancement improves the model's robustness and applicability for tasks requiring extensive context, as demonstrated by the provided Colab notebook test.

Highlights

• RoPE Embedding Extension: Implemented a mechanism to dynamically extend Rotary Position Embeddings (RoPE) to support longer sequence lengths within the Llama model.
• Long Sequence Training Fix: Addresses and resolves issues [Bug] AssertionError: seq_len > cos.shape[0] when dynamically extending RoPE beyond initial max_seq_length #3036 and [Bug] Assertion error in RoPE Embedding Kernel when training Qwen3-8B with long context #3216, which previously caused failures when training Llama models with extended sequence lengths.
• Llama Model Compatibility: Ensures the LlamaModel_fast_forward function correctly handles max_position_embeddings by proactively extending rotary_emb before retrieving cached position embeddings.

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 introduces a small but important fix for long sequence training in Llama models. By calling extend_rope_embedding within LlamaModel_fast_forward, it ensures that the Rotary Position Embedding (RoPE) cache is resized to accommodate the maximum configured sequence length. This proactive resizing prevents potential out-of-bounds errors when training with sequences longer than the default cache size, which is a common scenario in long-context finetuning. The change is correctly placed within the code path for newer transformers versions that use the attention refactor. The implementation is clean and directly addresses the reported issues. I approve of this change.