Feedback
It's pretty good! Can be quite lenghty, a little bit less than the previous one though, you can steer it more easily. But I like that this model has so much character compared to a lot of other models! You can really have some fun, specific characters with this one. Otherwise it has the usual 8B hallucinations, but overall, pretty impressive.
Thank you for the Amazing model, it's the best I have used till now even way better than 11B and 13B models I have used before but sometimes (only few times) it can get confused , I have been using Universal-light preset for sampler, any recommended preset that should be used?
also is it just me or this model is obsessed with thighs? specially my thighs XD
the model doesn't seem to be able to track the character and user's clothing.
the model doesn't seem to be able to track the character and user's clothing.
you should be happy if a model of this size was able to remember if your character is even clothed or not, for me I simply either fix it or ignore it.
the model doesn't seem to be able to track the character and user's clothing.
you should be happy if a model of this size was able to remember if your character is even clothed or not, for me I simply either fix it or ignore it.
Does a 'fix' in your webUI only change the text displayed to you on-screen, or also what the model remembers the conversation as having been?
@BingoBird - If you edit/fix the messages and save, yes, it changes the context, so it affects the entire "memory" of that conversation until that point and also going forward when that message is present in the context, if that's what you're asking.
I'm new to LLMs and this one I keep coming back to, so fantastic job. Do LLMs ever get updates? I'm just curious. This one is good for regular conversation and the NSFW stuff can get spicy. Thanks!!!
Do LLMs ever get updates?
Not in the same repo they don't.
The fact that model tuners do this is one of the reasons local AI is appealing: you don't have to worry about your setup breaking due to an update.
Sao10K did put out Stheno v3.4, but this is based on Llama 3.1, which can feel differently depending on what you liked. Honestly a lot of LLM updates are very subjective and not necessarily better in all areas, just enjoy what you enjoy and have a good time with your sessions.
I had uploaded quants on this repo:
https://huggingface.co/Lewdiculous/Llama-3.1-8B-Stheno-v3.4-GGUF-IQ-Imatrix
Another flavor based on L3.1 too:
https://huggingface.co/Lewdiculous/L3.1-8B-Niitama-v1.1-GGUF-IQ-Imatrix
Thanks for the answer, I'm new to all this and will keep learning. So far I'm impressed with LLMs and surprised at some of its shortcomings, but thats part of the process.
I had uploaded quants on this repo:
https://huggingface.co/Lewdiculous/Llama-3.1-8B-Stheno-v3.4-GGUF-IQ-ImatrixAnother flavor based on L3.1 too:
https://huggingface.co/Lewdiculous/L3.1-8B-Niitama-v1.1-GGUF-IQ-Imatrix
I would like to ask you to prepare Q4_0 versions of these models for mobile devices. I still use your quantization because the others did not deliver the same quality as you have provided for this model.
I would like to ask you to prepare Q4_0 versions of these models for mobile devices. I still use your quantization because the others did not deliver the same quality as you have provided for this model.
Heya, here you go, added the Q4_0 quants for these as you requested in these repos:
You're fantastic, thank you so much! I'm going to download these! Thank you very, very much!
What context and prompts are you guys using. On this model, I get a lot of this:
<|im_end|><|im_start|>user<|im_end|> <|im_start|>
When using Serene Pub with its default context.
Every so often, the prompt and context I use seem to cause the model to keep going for several turns.