Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Default String for ConfiguredSamplers #420

Open
JuliaMerz opened this issue Sep 4, 2023 · 1 comment
Open

Default String for ConfiguredSamplers #420

JuliaMerz opened this issue Sep 4, 2023 · 1 comment

Comments

@JuliaMerz
Copy link

JuliaMerz commented Sep 4, 2023

impl Default for ConfiguredSamplers {

We have a default for ConfiguredSamplers, as well as a FromStr implementation for users to give in a sampler string. However, at the moment we do not have that default represented anywhere. It would be useful to write the default string down somewhere so users can edit and use it themselves.

An LLM-based conversion of the default leads to this (which matches the structure in the default code correctly):

repetition:penalty=1.30:last_n=64 freqpresence:last_n=64 seqrepetition topk:k=40 tailfree locallytypical topp:p=0.95 temperature:temperature=0.8 mirostat1 mirostat2

However, the default (and therefore this string) contains both mirostat samplers, but sets mirostat1 and mirostat2 to false:

mirostat1: false,
mirostat2: false,
incompat_mirostat: false,

So there's something going on that I'm missing, or there's something off about the current default implementation.

@JuliaMerz
Copy link
Author

Alright, I finally dug into this properly. I got some debugs on the configured sampler, and was able to reconstruct our default implementation from a string.

-s "repetition:penalty=1.3:last_n=64 topk:k=40:min_keep=1 topp:p=0.95:min_keep=1 temperature:0.8"

creates the same configured sampler as our default.

I'll submit a small PR to throw this into a comment just for future users trying to learn about the code. I'll add it to the CLI help as well.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant