jackqdldd

Results 12 comments of jackqdldd

模型服务地址:http://xxx.com.cn:9316/multimodal/glm4v 参数:{ "query": "xxx", "image_path": "/data/ee_22_1714340162.jpeg", "history": [] } ![image](https://github.com/user-attachments/assets/a8a115fe-410e-4cd6-8605-16a662782690) 接口可以curl通,但是执行example_eval_vlm_swift.py 的时候一直卡着 2024-08-13 16:05:49,460 - evalscope - INFO - *** Run task with config: Arguments(data=['MMBench_TEST_CN'], model=['glm4v'], nframe=8, pack=False, use_subtitle=False, work_dir='output',...

![image](https://github.com/user-attachments/assets/0d2c2a84-4eca-4d3a-8db6-269a2e8acc57) python -m llmuses.run --model qwen/Qwen2-7B-Instruct --template-type qwen --datasets trivia_qa --limit 2

环境是一样的: modelscope Version: 1.16.0 llmuses 0.4.0 ![image](https://github.com/user-attachments/assets/428eba89-67bd-4409-ba10-d61ac0c60948) python llmuses/run.py --model qwen/Qwen2-7B-Instruct --template-type qwen --datasets arc --dataset-hub Local --dataset-args '{"arc": {"local_path": "/root/eval-scope/data/arc"}}' --dataset-dir /root/eval-scope/data/

果然是环境的问题,请问支持访问部署好的模型吗?比如远程部署了大模型,怎么通过地址来评测这个大模型

谢谢,我试了上面的方法是可以通的,不过这个是测性能的吧,对模型结果验证需要怎么做呢?用自带的数据集或者自定义数据集验证模型的能力,模型在远端机器部署着

> I have opened a PR to fix this. I have updated the docker-compose.yaml and the issue still exists: ![Image](https://github.com/user-attachments/assets/13d279ed-e494-490e-ad24-09ab5db04927) ![Image](https://github.com/user-attachments/assets/0604c3cf-b2c4-4b1a-b15f-9dc4b8db04c2)