search

cocktailpeanut / dalai

The simplest way to run LLaMA on your local machine
https://cocktailpeanut.github.io/dalai
13.09k stars 1.42k forks source link

Models repository #470

Closed redskate closed 10 months ago

redskate commented 11 months ago

Dear community!

Thanks a lot for his very interesting system!!! However: The only interesting model here is the 65B and this happens to be huge and the only one mentioned and shown in the article https://arxiv.org/abs/2302.13971 (practically the only one of interest).

I managed to download 7B, 13B and 30B and they work "nice" but not really useful in answering (see your prompts) - (thank you).

Installing the 65B model resulted for me since 4 days in a "try" to download the consolidated.xx.pth files (each day a new try). This npx install blocks after having downloaded 3 "consolidated" files ... and I have no idea what might have gone wrong here and why (and I stop the download, but then all the downloaded files are erased to start again from scratch). This should not be (or yes?)

Is there a URL on a repository where one can directly manually download what is shown in https://github.com/cocktailpeanut/dalai/blob/main/docs/65b.png ???

Or else what do you suggest please?

Thanks a lot! Kind regards

redskate commented 10 months ago

And here my comment (since I observe, no reaction has come so far). Thank you again for all the llama models! I managed to launch a docker version from a good connected server and could download + quantize all the (heavier) llama models.

Of course a llama 65B model takes too long to load (5-10min on an average machine at each use) and an alpaca 65B model would be faster and very appreciated but is not available.

I close therefore this issue, hoping at some point to read something on the alpaca 65B.

Good work.