Model Series

Qwen3

What are the models used?

Qwen-3-14B, Qwen-3-14B-AWQ, Qwen3-235B-A22B

What is the scenario where the problem happened?

Qwen 3 starts repeating emojis infinitely when asked to list them, both with vLLM and on chat.qwen.ai

Is this badcase known and can it be solved using avaiable techniques?

• I have followed the GitHub README.
• I have checked the Qwen documentation and cannot find a solution there.
• I have checked the documentation of the related framework and cannot find useful information.
• I have searched the issues and there is not a similar one.

Information about environment

Replicated on chat.qwen.ai with Qwen-3-235B-A22B, but also

Qwen-3-14B, 4xA10g 24GB, vLLM v0.8.5, Ubuntu 22.04, CUDA 12.04

and

Qwen-3-14B-AWQ, 1xA100 40GB SXM4, vLLM v0.9.2, Ubuntu 22.04, CUDA 12.04

Description

Steps to reproduce

This happens to Qwen3-14B and Qwen3-14B-AWQ hosted locally with vLLM and also Qwen3-235B-A22B on chat.qwen.ai

The problem can be reproduced with the following steps:

1. Ask it to list every emoji it knows
2. Wait for the output to start repeating

chat.qwen.ai screenshot:

vLLM screenshot:

Expected results

It doesn't loop and stops the generation after a normal amount of time

Attempts to fix

N/A

Anything else helpful for investigation

N/A