https://huggingface.co/Elfrino/PsyMedRP_XL_v2

#316
by Elfrino - opened

Could I please get Imat Quants of this model?

Thankyou in advance and keep up the fantastic work! :)

Heh, just queued it myself. I'll bump it's priority then :)

You can watch it progress at http://hf.tst.eu/status.html

mradermacher changed discussion status to closed

Thankyou!

This is a highly experimental merge, really want to beef up the size of PsyMedRP.

Hoping downloading a few different Quant sizes and testing them will let me see if it's worth pursuing this mad experiment! :)

psymedrp was truly an interesting model, hopefully something comes out of it :)

Alas after more testing unfortunately this merge was not a success. I thought it had that PsymedRP creative spark at the beginning but no, it is somewhat 'drunk' and incoherent later on in the stories.
Drat, I really love PsyMedRP's creativity and a bigger version of it that is as coherent and creative (hopefully moreso) then the original is my goal.

I apologize for my mistake, please delete this model if you like. I don't want it cluttering your awesome collection and wasting people's time and energy if they choose to download it.

In regards to PsyMedRP it seems like a normal model until you set specific parameters and then it's wacky, bizzare and scifi creativity comes through:

--KoboldCPP settings--

Quick preset: Mayday
Temp: 1.3
Top P sampling: 0.99
Repition penalty (between 1.08 and 1.1)
Prompt template: Alpaca

(works great with PsymedRP imat version, around i1-Q5_K_M, i1-Q4_K_M)

I will keep tinkering with trying to merge this model and hopefully get a bigger and better version of it. Thanks again. :)

Believe me, I have a lot worse models in my collection than "could be better". So don't worry, it's totally fine to request a quant even if the model is unproven or turns out to not be the next top model... :)

So don't hold back in the future. It costs us much less to quantize a model than to test it or create it.

Sign up or log in to comment