LensonYuan

Results 12 comments of LensonYuan

The author who currently provides script , almost written not accurate

![Uploading image.png…]()

terminal do "which stats", then, like this, "type": "appleScriptTitledButton", "source": { "inline":"set temp to do shell script \"/usr/local/bin/istats cpu --no-graphs --no-labels\"\r return temp " },

please see ckpt/readme.txt and download, using 'tar -zxvf name.tar.gz', then copy to ckpt

It should be a problem with the TensorFlow version. Check that the version is TF 1.0.0

我这边用了faiss向量库和其他embeding模型算法。所以截取不到代码了,大致修改如下: ```python if self.num_expand_context_chunk > 0: new_reference_results = [] for corpus_id, hit_chunk in hit_chunk_dict.items(): expanded_reference = self.sim_model.corpus.get(corpus_id - 1, '') + hit_chunk for i in range(0, self.num_expand_context_chunk+1, 1): expanded_reference +=...

另外,当文章内容很少时,这个机制会导致Prompt的内容大量重复,比如基于corpus_id=3去拉取上下文,有可能刚好拉取到了已经在相关列表中的chunks. 那这样提供的prompt正文参考,会出现大量重复的chunk。

![image](https://github.com/user-attachments/assets/82968827-8082-4b36-b051-337370310ae4) 这个位置代码也有问题,建议重构下切分逻辑。中英文符号是很难区分开的。另外一个是,截框里的条件判断有问题,not current_chunk一定为true的。还有sentences分的太长超过了chunk_size,也照样返回了,这会导致后续逻辑embedding异常,维度超表了。

> 登录接口请求参数需要是表单类型,看看你前端是不是写错了 是的,前端还没改,我想调试中去改,能否请教下断点应该设置在哪里才有效呀