Fix AttributeError for supports_tower_connector_lora in vLLM 0.14+

[danielhanchen]

Summary

• vLLM 0.14.0 (PR #26674) added supports_tower_connector_lora() to WorkerLoRAManager for multimodal LoRA on vision encoder/connector layers
• gpu_model_runner.py calls it unconditionally for any LoRA-enabled model:

  if self.lora_config and self.lora_manager.supports_tower_connector_lora():

• Our patched WorkerLoRAManager (replaced via patch_vllm_lora_load_tensors) did not define this method, causing AttributeError when fast_generate is called on any LoRA-enabled VLM (e.g. Qwen2.5-VL GRPO)

Fix

Add supports_tower_connector_lora() to WorkerLoRAManager using getattr with safe defaults. LRUCacheWorkerLoRAManager inherits it automatically, matching vLLM's class hierarchy.

Uses getattr instead of direct attribute access because:

• _adapter_manager is lazily initialized by create_lora_manager() and may not exist yet
• Pre-v0.14 LoRAModelManager does not have supports_mm or supports_tower_connector_lora
• In v0.14+ non-multimodal models, _maybe_init_mm() is never called so supports_tower_connector_lora does not exist on the manager (vLLM's canonical code relies on and short-circuit; our getattr is strictly safer)

Compatibility

Tested across all vLLM version scenarios from v0.8.5post1 through v0.17.1:

Scenario	Result
_adapter_manager not yet set (pre-create_lora_manager)	False
vLLM 0.8.5-0.12.x (no supports_mm, no supports_tower_connector_lora on manager)	False
vLLM 0.9-0.13 multimodal (supports_mm=True, no supports_tower_connector_lora attr)	False
vLLM 0.14+ non-multimodal (supports_mm=False, no tower attr)	False
vLLM 0.14+ multimodal, enable_tower_connector_lora=False (default)	False
vLLM 0.14+ multimodal, enable_tower_connector_lora=True, model supports it	True
LRUCacheWorkerLoRAManager inherits correctly in all cases	Confirmed

[gemini-code-assist]

Summary of Changes

Hello, 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 addresses a critical AttributeError that prevented LoRA-enabled multimodal models from functioning correctly with vLLM versions 0.14.0 and newer. By safely introducing the supports_tower_connector_lora method to the custom LoRA manager, it ensures robust compatibility across a wide range of vLLM versions, preventing runtime crashes and enabling seamless operation of multimodal LoRA features.

Highlights

• Fixes AttributeError: Resolved an AttributeError that occurred when fast_generate was called on LoRA-enabled multimodal models due to a missing method.
• Adds supports_tower_connector_lora: Implemented the supports_tower_connector_lora() method in WorkerLoRAManager to align with changes introduced in vLLM 0.14.0 and newer.
• Ensures Backward Compatibility: Utilized getattr for safe access to attributes, ensuring compatibility across various vLLM versions and gracefully handling cases where managers or attributes might not be initialized or present.

Changelog

• unsloth_zoo/vllm_lora_worker_manager.py
  • Added supports_tower_connector_lora method to handle multimodal LoRA compatibility checks.

Activity

• No human activity has been recorded for this pull request yet.

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 addresses an AttributeError for supports_tower_connector_lora in vLLM 0.14+ by adding the missing method to WorkerLoRAManager. The implementation is robust, using getattr to ensure compatibility across different vLLM versions and initialization states. My review includes a minor suggestion to improve code conciseness.

[gemini-code-assist]

For conciseness, you can combine the null check for manager with the subsequent boolean logic into a single return statement. This avoids the explicit if/else flow and makes the function's logic more direct.

        manager = getattr(self, '_adapter_manager', None)
        return (
            manager is not None
            and getattr(manager, 'supports_mm', False)
            and getattr(manager, 'supports_tower_connector_lora', False)
        )