bubble

Results 7 issues of bubble

您好,我发现第七章代码中有处地方能够优化一下。 tokenizer函数中,可以去掉padding='max_length',浪费计算资源。transformer提供的Trainer构造时的data_collator参数默认采用了动态补全的方法,按照batch进行补全,能够节省计算资源。 在我的CPU上跑,时间从36小时变为2小时(没跑完,进度条给的预估时间)

I noticed the annotation ```python # If you're using Seqio, we suggest caching your mixture as they take a while to generate. ``` But I don't know how to do...

请问[make_data_example.py](https://github.com/ssbuild/chatglm_finetuning/blob/dev/data/make_data_example.py)中将数据重复100次, 是想要研究多少数据才能让模型记住这个例子吗? 有相关的结论吗?

以及 judical_examination.json | 2,000 | ChatGPT生成的法考题解答 -- | -- | -- judical_examination_v2.json | 5,000 | ChatGPT生成的法考题解答(第二批公开) 这两份数据是使用JE-Q2EA还是JE-QA2E方法构造的呀 ? 有什么其他的处理吗 ?

The tick ✔️ means that the corresponding corpus/dataset has been used at the previous stage, while the flower 🌸 means the corpus/dataset is employed for training at the current stage....

What is the processing of the inference and extraction for humaneval? My test score is only 20.73 for humaneval (k=1, model = CodeLlama-7b-Instruct-hf+DPO ).

请问用于代码修复能力评估的prompt是啥呀