Alex "mcmonkey" Goodwin

See conversation @ https://github.com/oobabooga/text-generation-webui/pull/1098#issuecomment-1509713861 tldr: looks like current version of peft is bugged with file saving. Using `pin install peft==0.2.0` to go back might fix it. Alternately, if you use...

This will be fixed when #1098 is pulled - if impatient you can pull from that PR directly to use.

Adding `dolly` to that matcher indeed fixes the `--verbose` output to contain the text properly. However while testing it generated: ``` ### End! More unwanted text here ``` Despite having...

Oo, that PR is a really good improvement. Definitely a good way to fix the dolly issue too. ... Is that config automatically applied in the UI? ie, can a...

The primary distinction from Alpaca format is it has `### End` as a dedicated stop token. In my earlier tests it output the token consistently but having it as stopping...

I can confirm aside from a typo I commented on that, and having my own Dolly file (Alpaca + stop string) and model config, that PR mostly works. Only limitation...

Indeed appears to be working perfectly now! And right out of the box too, all automated, I love it.

Parameters should apply instantly. Note that some parameters are conditional whether they apply (eg `top_p` and `top_k` don't really co-exist) so changes don't do much. If you want to confirm...

Not all concepts quite so directly map between SD and LLMs. For example, re `a prompt alternating by changing the input prompt as shown at each timestep that the LLM...

I apologize for my earlier post assuming you were a new/inexperienced user that thought LLMs and SD worked the same in their inputs. Your phrasing (eg you mentioned `timesteps` which...