Your text-only requests work, but image requests to Qwen/Qwen2.5-VL-3B-Instruct return “IndexError: list index out of range” because this is a known bug in vLLM 0.10.x when handling some image inputs, especially with certain image URLs or formats. This error is reproducible and has been reported by other users as well, often when using the OpenAI-compatible API with image content for Qwen2.5-VL models. The issue is not with your request format, which is correct per the OpenAI Vision API spec supported by vLLM.
A workaround is to downgrade to vLLM 0.10.0 (not 0.10.1+) or use a different image URL, as some users report the error disappears with certain images or vLLM versions. If you must use 0.10.1+, you may need to wait for an upstream fix or try serving with VLLM_USE_V1=1 (though this is not always effective). For more, see this issue and this example.
Would you like a step-by-step troubleshooting guide or example code for a working image request?
Sources: