Support emb/reranker/seq_cls padding_free

[tastelikefeet]

PR type

• Bug Fix
• New Feature
• Document Updates
• More Models or Datasets Support

PR information

Write the detail information belongs to this PR.

Experiment results

Paste your experiment result here(if needed).

[gemini-code-assist]

Summary of Changes

Hello @tastelikefeet, 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 implements a significant new feature to enable padding-free training for various natural language processing tasks, specifically embedding, reranking, and sequence classification. The core idea is to optimize memory and computation by avoiding explicit padding during the model's forward pass for variable-length inputs. This is achieved through extensive refactoring of model patching mechanisms and the introduction of utilities to handle sequence length management and dynamic re-padding, ensuring that models can operate efficiently with unpadded data while maintaining compatibility with existing frameworks.

Highlights

• Padding-Free Support: Introduced support for padding-free training across embedding, reranker, and sequence classification tasks, enhancing efficiency by processing variable-length sequences without explicit padding during the core model forward pass.
• Model Patching Refinements: Refactored model patching logic in swift/llm/model/patcher.py and swift/trainers/mixin.py to dynamically modify model forward passes for padding-free operation, including unwrapping PeftModel instances and handling SentenceTransformer specific architectures.
• Sequence Classification Logic: Extracted and enhanced sequence classification pooling logic into a new transformers_seq_cls_forward function, which correctly determines sequence lengths for pooling based on attention_mask or input_ids.
• Dynamic Re-padding Mechanism: Added a revert_padding_free utility function that reconstructs padded outputs and attention masks from padding-free model outputs, ensuring compatibility with subsequent operations or standard data formats.
• Training Script Updates: Updated several reranker training scripts (train_generative_reranker.sh, train_reranker.sh, etc.) to include the --padding_side left argument, aligning with the new padding-free capabilities.
• Data Handling for Rerankers: Modified the encode method in swift/llm/template/base.py to store sequence lengths as a list for reranker and generative reranker tasks, accommodating the variable-length nature of padding-free processing.

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 support for padding-free training for embedding, reranker, and sequence classification tasks, which is a great performance enhancement. The core logic involves patching the model's forward pass to handle packed sequences and then reverting to a padded format for downstream components. The implementation is well-structured, centralizing the patching logic in _patch_tasks.

My review includes a few suggestions to improve robustness and maintainability:

• Handling potential errors when required keys are missing in inputs.
• Refactoring duplicated code blocks to improve readability.
• Addressing a potential bug in fallback logic for sequence length calculation.
• Fixing a bug in a utility function that could cause a crash with empty list inputs.

Overall, this is a solid contribution. Addressing these points will make the new feature more robust.