Tianxiang Sun

First check if your `transformers` version is 4.1.1, or you can simply remove `@add_code_sample_docstrings` (line 631-636)

No, our prompt is continuous and it may hurt performance if rounding it into a discrete prompt. I'd like to recommend some related works that optimize discrete prompts: [https://github.com/txsun1997/LMaaS-Papers#black-box-optimization](https://github.com/txsun1997/LMaaS-Papers#black-box-optimization)

Hi @jordane95 We've updated our code and results recently. Please check the latest implementation (mainly in `bbt.py` and `deepbbt.py`). New results can be found in [Google Sheets](https://docs.google.com/spreadsheets/d/1FA9zMW613OoskI_fBXuZNNBVo7RFV5OvdZ0YHnA2j5Q/edit?usp=sharing), where we list...

@SnakeHacker meta instruction格式请参考我们的[示例数据](https://github.com/OpenLMLab/MOSS/tree/main/SFT_data/conversations/conversation_with_plugins/calculator)的格式以及[README插件增强部分](https://github.com/OpenLMLab/MOSS#%E6%8F%92%E4%BB%B6%E5%A2%9E%E5%BC%BA)。 MOSS仅生成插件执行命令，即`Calculate(12+25)`，您需要自行实现插件的执行过程，并将结果按照[示例数据](https://github.com/OpenLMLab/MOSS/tree/main/SFT_data/conversations/conversation_with_plugins/calculator)的格式放在Results中

@Ratuchetp @ajz34 Hi 目前量化版本模型仅支持单卡部署。请不要使用`model.half().cuda(1)`或者`inputs[k].cuda(1)`，而是CUDA_VISIBLE_DEVICES=1 python xxx.py，在python代码里使用`model.half().cuda()`和`inputs[k].cuda()`

请检查环境依赖是否对应，我复现没出现问题

@Fu-Dayuan 可以直接输出答案，比如在拿到输入后直接拼接一个 ``` : 你好 : None : None : None : ``` 输入给模型。

@ImGoodBai https://github.com/OpenLMLab/MOSS#%E6%A8%A1%E5%9E%8B%E9%87%8F%E5%8C%96

https://github.com/OpenLMLab/MOSS/issues/14

@xerrors 插件版MOSS的使用说明已经更新到README