[Bugfix] Fix Qwen2 audio chat template for old version transformers compatibility #20826

Isotr0py · 2025-07-11T16:05:24Z

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

Fix [Bug]: Qwen2Audio 500 internal Error #20796
Although Transformers has fixed the Qwen2-Audio in new version, we still need to replace the chat template for old version Transformers installation

Test Plan

Test Result

(Optional) Documentation Update

Signed-off-by: Isotr0py <2037008807@qq.com>

github-actions · 2025-07-11T16:05:33Z

👋 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 @Isotr0py, 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 provides a bugfix for Qwen2-Audio models, specifically addressing compatibility with older versions of the transformers library. It introduces a new, explicit chat template for Qwen2-Audio and enhances the chat template resolution system to allow for overriding default templates with predefined fallbacks, ensuring correct prompt formatting across different transformers versions.

Highlights

Qwen2-Audio Chat Template Fix: A new Jinja template (template_qwen2_audio.jinja) has been added specifically for Qwen2-Audio models. This template ensures proper formatting of both text and audio inputs, addressing compatibility issues with older transformers library versions.
Enhanced Chat Template Resolution Logic: The system for resolving chat templates has been improved. It now introduces a TemplateInfo dataclass with an override_exists flag, allowing predefined fallback templates to explicitly override templates obtained from AutoProcessor or AutoTokenizer when necessary. This ensures that models like Qwen2-Audio use the correct template even if the default one is problematic.
Refactored Chat Template Fallback Mechanism: The internal get_chat_template_fallback and get_chat_template_fallback_path functions have been updated to leverage the new TemplateInfo structure, providing a more robust and flexible way to manage and apply model-specific chat templates, especially for compatibility purposes.

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.

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 mechanism to override Hugging Face chat templates, specifically to fix an issue with Qwen2-Audio on older transformers versions. The approach is solid and well-implemented. I've provided a few suggestions to enhance robustness and code clarity, including handling a potential None value that could cause a runtime error and correcting a condition in the new Jinja template.

gemini-code-assist · 2025-07-11T16:07:07Z

vllm/entrypoints/chat_utils.py

+    if path is not None and chat_template is None:
        logger.info("Loading chat template fallback for %s as there isn't one "
                    "defined on HF Hub.", tokenizer.name_or_path)
        chat_template = load_chat_template(path)
+    elif fallback_info is not None and fallback_info.override_exists:
+        chat_template_to_override = load_chat_template(path)
+        if chat_template_to_override != chat_template:
+            chat_template = chat_template_to_override
+            logger.warning("Override existing chat template for %s as "
+                           "its defined ones on HF Hub is not compatible "
+                           "with OpenAI server.",
+                            tokenizer.name_or_path)


To prevent a potential TypeError, ensure path is not None before calling load_chat_template(path). If fallback_info.get_path() returns None, load_chat_template(None) would raise an error.

Suggested change

if path is not None and chat_template is None:

logger.info("Loading chat template fallback for %s as there isn't one "

"defined on HF Hub.", tokenizer.name_or_path)

chat_template = load_chat_template(path)

elif fallback_info is not None and fallback_info.override_exists:

chat_template_to_override = load_chat_template(path)

if chat_template_to_override != chat_template:

chat_template = chat_template_to_override

logger.warning("Override existing chat template for %s as "

"its defined ones on HF Hub is not compatible "

"with OpenAI server.",

tokenizer.name_or_path)

# 4th priority: Predefined fallbacks

fallback_info = get_chat_template_fallback(

model_type=model_config.hf_config.model_type,

)

path = (fallback_info.get_path(model_config.tokenizer)

if fallback_info is not None else None)

if path is not None and chat_template is None:

logger.info("Loading chat template fallback for %s as there isn't one "

"defined on HF Hub.", tokenizer.name_or_path)

chat_template = load_chat_template(path)

elif fallback_info is not None and fallback_info.override_exists and path is not None:

chat_template_to_override = load_chat_template(path)

if chat_template_to_override != chat_template:

chat_template = chat_template_to_override

logger.warning("Override existing chat template for %s as "

"its defined ones on HF Hub is not compatible "

"with OpenAI server.",

tokenizer.name_or_path)

gemini-code-assist · 2025-07-11T16:07:07Z

vllm/transformers_utils/chat_templates/template_qwen2_audio.jinja

+        {{ message['content'] + '<|im_end|>\n' }}
+    {%- else -%}
+        {%- for content in message['content'] -%}
+            {%- if 'audio' in content or 'audio_url' in content or message['type'] == 'audio' or content['type'] == 'audio' -%}


The condition message['type'] == 'audio' is likely incorrect, as message refers to the entire message object (e.g., {'role': 'user', 'content': [...]}) which typically doesn't have a type key. The check should be on the content part. Also, the check content['type'] == 'audio' is duplicated. A more correct and robust condition would be to check for content.get('type') == 'audio' and remove the other checks.

{%- if 'audio' in content or 'audio_url' in content or content.get('type') == 'audio' -%}

gemini-code-assist · 2025-07-11T16:07:08Z

vllm/transformers_utils/chat_templates/registry.py

+    chat_template_info = get_chat_template_fallback(model_type)

-    if chat_template is None:
+    if chat_template_info is None:
        return None

-    return chat_template
+    return chat_template_info.get_path(tokenizer_name_or_path)


Consider simplifying this function by directly returning the result of get_chat_template_fallback(model_type). This eliminates the need for the chat_template_info variable and the subsequent conditional checks, making the code more concise.

Suggested change

chat_template_info = get_chat_template_fallback(model_type)

if chat_template is None:

if chat_template_info is None:

return None

return chat_template

return chat_template_info.get_path(tokenizer_name_or_path)

def get_chat_template_fallback_path(

model_type: str,

tokenizer_name_or_path: str,

) -> Optional[Path]:

return get_chat_template_fallback(model_type).get_path(tokenizer_name_or_path) if get_chat_template_fallback(model_type) else None

DarkLight1337

Can merge once the author of the original issue confirms this fix

Isotr0py added 3 commits July 11, 2025 23:14

fix qwen2-audio backward compatibility

9bff87c

Signed-off-by: Isotr0py <2037008807@qq.com>

fix

3b40a27

Signed-off-by: Isotr0py <2037008807@qq.com>

fix chat template

37a35cc

Signed-off-by: Isotr0py <2037008807@qq.com>

Isotr0py requested a review from DarkLight1337 July 11, 2025 16:05

Isotr0py requested a review from aarnphm as a code owner July 11, 2025 16:05

gemini-code-assist bot reviewed Jul 11, 2025

View reviewed changes

mergify bot added frontend qwen Related to Qwen models labels Jul 11, 2025

Isotr0py mentioned this pull request Jul 11, 2025

[Bug]: Qwen2Audio 500 internal Error #20796

Open

1 task

gemini-code-assist bot reviewed Jul 11, 2025

View reviewed changes

DarkLight1337 approved these changes Jul 11, 2025

View reviewed changes

DarkLight1337 added the ready ONLY add when PR is ready to merge/full CI is needed label Jul 14, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

[Bugfix] Fix Qwen2 audio chat template for old version transformers compatibility #20826

[Bugfix] Fix Qwen2 audio chat template for old version transformers compatibility #20826

Isotr0py commented Jul 11, 2025 •

edited by github-actions bot

Loading

Uh oh!

github-actions bot commented Jul 11, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Jul 11, 2025

Uh oh!

gemini-code-assist bot Jul 11, 2025

Uh oh!

gemini-code-assist bot Jul 11, 2025

Uh oh!

DarkLight1337 left a comment

Uh oh!

Uh oh!

Uh oh!

[Bugfix] Fix Qwen2 audio chat template for old version transformers compatibility #20826

Are you sure you want to change the base?

[Bugfix] Fix Qwen2 audio chat template for old version transformers compatibility #20826

Conversation

Isotr0py commented Jul 11, 2025 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Essential Elements of an Effective PR Description Checklist

Purpose

Test Plan

Test Result

(Optional) Documentation Update

Uh oh!

github-actions bot commented Jul 11, 2025

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

DarkLight1337 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Isotr0py commented Jul 11, 2025 •

edited by github-actions bot

Loading