Checklist
Describe the bug
If I run the Qwen2.5-VL-7B-Instruct model with dynamic quantization using SGLang (--quantization w8a8_fp8 or fp8), the server crashes with an error when receiving a request containing an image.
Logs:


Reproduction

Environment
sglang docker v.0.4.6.post5
Checklist
Describe the bug
If I run the Qwen2.5-VL-7B-Instruct model with dynamic quantization using SGLang (--quantization w8a8_fp8 or fp8), the server crashes with an error when receiving a request containing an image.
Logs:
Reproduction
Environment
sglang docker v.0.4.6.post5