liushz
liushz
Primarily, our decoding phase employs greedy search, and fixed generation parameters (like temperature and so on) for every model ensuring theoretical consistency in the responses generated. However, there are nuances...
> > It seems like the right version provided a structural description for the triangle, the original problem and answer are same. > > On other problems, there are also...
感谢您的建议，我们会尽快测试该模型并支持，正常的话eval阶段vllm也是没问题的，你可以试下用其他模型在eval阶段用vllm推理，可能是这个模型跟OpenCompass的vllm不适配，我们这边也会尽快验证这个事情。
Fixed in https://github.com/open-compass/opencompass/pull/1430
Due to a recent update in OpenCompass, certain datasets for `./configs/eval_gpt3.5.py` are no longer supported. If you wish to test GPT3.5 using a demonstration configuration, please navigate to `./configs/eval_gpt3.5.py` and...
Which section of the documentation allows you to use this configuration for the demo? `./configs/eval_gpt3.5.py`
Which function do you mean?
Can you show some responses without the `pred_postprocessor`.
When your model type is HF like `HuggingFacewithChatTemplate`, do not use `gen_config` to pass the model parameters, use `model_kwargs` instead.
Please see the complete prediction in `predictions` dir