wluo7

Qwen2.5-VL-32B-Instruct fp16 image recognition issue

1

**Describe the bug** using intelanalytics/ipex-llm-serving-xpu:0.8.3-b18 to serve Qwen2.5-VL-32B-Instruct, setting low bit to fp16 will not return image discription, while setting low bit to fp8 works fine. **How to reproduce** Steps...

intelanalytics/multi-arc-serving:0.8.3-b21 on CORE i7 13900 * A770 stress testing fail causing system dead after serval rounds

1

**Describe the bug** intelanalytics/multi-arc-serving:0.8.3-b21 on 2*A770 Core i7 13900, batch size 20, input token 1024, output token 512, QwQ-32B-AWQ. stress test fail after several rounds, ssh connection to the server...