-
微博内容精选
-
你好！
请教一下咱们训练自有中文数据的流程是怎样的？
我的理解是首先要分词，也就是 tokenizer . 但是我看你的介绍用 sentensepiece，这个是分词工具吗？适合中文吗？
方便介绍一下您是怎么从头训练中文语料的吗？
谢谢了!
-
Did u consider any research paper for this project? If yes, please let me know.
How did you calculate the accuracy of this project?
-
I kept getting out-of-memory errors (CUDA) with PCGrad until I put the batch size all the way down to 16.
This might be related to the fact that we don't support AMP with PCGrad right now: https://gi…
-
-
PCGrad (Gradient surgery) requires the loss for each task, which assumes you ran 1 batch per task, which requires round-robin scheduling. How could we integrate other schedulers into that? (especially…
-
-
-
**Describe the bug**
Using current main branch (without any change in the code), several test cases fail
**To Reproduce**
Steps to reproduce the behavior:
1. Clone the project to your local mac…
-
# Rationale?
[집현전에서 발표](https://youtu.be/AvW8zX9xG8s)를 하면서, 명준님의 이 논문을 정말 재미있게 읽었다 |
--- |
|
논문의 제목이 Let Language Models Learn Meaning-Text Correspondence인데. 명준님과 미리 이야기를 해보기도 했어서, 논문의 내용에 …