wallon-ai
wallon-ai
> is everyone still having this issue? I keep getting the same interruption "Stream interrupted (client disconnected).", i'm hoping that its just the stream that is interrupted and not the...
+1
> @better629 My inference is also slow, though I only use a single RTX-8000 GPU. I even load the model using `load_in_8bit=True`. The inference takes around 12s for `max_new_tokens=64` and...
same issue. I'm on ubuntu.
> You ran out of GPU memory. Describe more on your setup like what you are using and what command you ran to resolve.  batch_size=4
我也遇到了相同的问题，请问有解决办法了吗
你好，文档稍微长一点时，文档后面的内容模型好像学不到，请问你知道怎么解决这个问题吗？
但是gpt4还是有长度限制，就特别好奇chatpdf是如何解决这个问题的？ ------------------ 原始邮件 ------------------ 发件人: ***@***.***>; 发送时间: 2023年3月28日(星期二) 下午3:11 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [fierceX/Document_QA] 为什么不直接用Embedding召回的结果呢？ (Issue #1) @wallon-ai 最好的办法应该是gpt4或者针对性的微调，但是微调的代价应该很大（openai 还没放出来微调api），gpt4拥有更长的上下文，这样就能解决长文档的问题。 — Reply to this email directly, view it on...