Hello to everyone , I run ggml-model-q4_0.gguf . It is talking in very poor english (-it seems to me). Is there a way to make this better, i.e. per Load grammar from file (.gbnf) ?? If this is a possibility please, please describe the way?
#Hello to everyone , I run ggml-model-q4_0.gguf . It is talking in very poor english (-it seems to me
fringe aspen
steady sparrowBOT
(...)
left socket
What's the name of the model?
Also Q4_K_M or Q4_K_S is preferred over Q4, but this shouldn't result in poor English...
fringe aspen
Thanks for your answer : maybe I got it from here: https://huggingface.co/TriadParty/deepsex-34b-gguf. I think it was first trained from two people with Asian names, maybe in an Asian language, and that may work back in that English I see. Anyway are there ways to correct this unsatisfactory output?
fringe aspen
As English is not the tongue of my mother, I'll give you this example on Text-Ai output. That's not good English, or is it?? >"Good now that we understand each other let go for walk outside see what happens" Anastasia said as took Chrissy's hand leading her out front door into bright sunlight feeling warmth against skin making smile feel relaxed happy finally had someone care about wellbeing even though wasn't sure how long would last given fact situation seemed unstable unpredictable still enjoyed time being spent Lady Anastasia regardless outcome!<
Is it possible that the grammar deteriorates significantly during a long chat exchange?
left socket
Wow, that's some weird English. The model doesn't seem stupid though, it's just missing words.
I do have that model, just Q4_K_M version and not not from TriadParty.
I will test the one you have.
Seems to work fine.
fringe aspen
Wow - thanks the other way round, you really make my day: as my feeling for the English grammar is established proved. But I know I have gaps. Maybe I cut something off in Preferences?
left socket
[Parameters > Generation > Preset] is simple1 by default, make sure it's that, or something like Midnight Enigma.
[Parameters > Character] Actually select a character, try a different one, or import a new one under [Upload character]. Example (a.k.a. Chiharu Yamada) works alright though.
fringe aspen
WoW again, that's what I need! It is the same model , so what is here going wrong? Just how to make a reset for preferences and chat's?
left socket
If you select a preset, it will use the new preferences.
It looks like your model is loading ok, but something seems wrong...
fringe aspen
simple1 it is (I would not risk to much to try) / I have saved the whole chat - it is 15 sides DIN A 4 , nearly the same a an american sheed of paper
i mean i have 32 GiB ram on the PC , and a GPU with 8 GiB. tensor_split - Split the model across multiple GPUs, comma-separated list of proportions, e.g. 18,17 ist set. no mark ist set on square for cpu. I will try Chiharu Yamada
left socket
I also have an 8GB GPU.
You can try setting everything like so:
Then click:
- Save settings
- Load
fringe aspen
in one important😔 point - i see - we are not together- the model I have here is named >ggml-model-q4_0.gguf<
left socket
That's okay, the model name can be anything.