Hao Zhang
Hao Zhang
Are you saying L2 error? Hao On Sun, May 21, 2017 at 4:31 AM, ningning32 wrote: > How to calculate L^2 error? > > — > You are receiving this...
It seems the training speed with Deepspeed isn't great. We'll add some better model-parallel training support soon. Closing this ticket.
Unfortunately we're unable to help on this issue. @eeric maybe try to do some search on hugging face?
Is the issue resolved?
It can take from 1-2 days to 2 months, based on my experience.
You can write a simple throughput calculator here https://github.com/lm-sys/FastChat/blob/main/fastchat/serve/inference.py#L98 and estimate the throughput (e.g., words/s) right? Contributions are welcome.
It seems the problem of this issue has resolved -- closing.
Closing, as the issue has been resolved.
Please use the Vicuna 1.1 new weight delta and new apply_delta script, which shouldn't have any issue. Feel free to re-open if you find any issue!
yes, 7B is worse than 13B