Lewdiculous

/

Lumimaid-v0.2-8B-GGUF-IQ-Imatrix

like 29

Damn, that's the sauce! I'm already testing this out, liking it so far!

Good stuff. I'll upload the 12B Nemo Based too since I'm personally curious about it as well.

I'm getting this error on the latest koboldcpp

llama_model_load: error loading model: done_getting_tensors: wrong number of tensors; expected 292, got 291
llama_load_model_from_file: failed to load model

@xalkoi - Make sure you're in the latest version, that being 1.71.1, not 1.71. The issue you're reporting seems to be been this one, which was fixed via a .1 hotfix.

If you installed koboldcpp using Scoop.sh, use the scoop update --all command.

If that really doesn't work, can you test with the latest cuda build from here?

Situation:

KoboldCpp 1.71 = Error
KoboldCpp 1.71.1 = Works

@xalkoi - Make sure you're in the latest version, that being 1.71.1, not 1.71. The issue you're reporting seems to be been this one, which was fixed via a .1 hotfix.

If you installed koboldcpp using Scoop.sh, use the scoop update --all command.

If that really doesn't work, can you test with the latest cuda build from here?

---

Situation:

KoboldCpp 1.71 = Error
KoboldCpp 1.71.1 = Works

I had to uninstall and re-install my koboldcpp through scoop, it was freaking out because the hashes are different from 1.71 and 1.71.1 but seemed to be in the same spot
Ps: 64K ctx is possible with 4bit cache in 8gb x.x

Getting this error in Oobabooga WebUI, latest version. Is there a way to fix this?

Any recommended sampler preset? because this model is too horny, it takes imitative too fast and without any reason.

Any recommended sampler preset? because this model is too horny, it takes imitative too fast and without any reason.

Taking the advice Grimjim gave to someone in the SillyTavern discord, you could write a message asking why the character is so horny with an OOC prompt, then use the response to help with creating a reverse jailbreak.
Here's an example:

Why is the character overly sexual?

[OOC:
Pause the role-play.
You are now the AI assistant, designed to help {{user}} understand aspects of the role-play and what the characters are thinking.]

Any recommended sampler preset? because this model is too horny, it takes imitative too fast and without any reason.

Taking the advice Grimjim gave to someone in the SillyTavern discord, you could write a message asking why the character is so horny with an OOC prompt, then use the response to help with creating a reverse jailbreak.
Here's an example:

Why is the character overly sexual?

[OOC:
Pause the role-play.
You are now the AI assistant, designed to help {{user}} understand aspects of the role-play and what the characters are thinking.]

I will give it a try, thank you

I love the way this model feels and acts. But I seem to have an Extremely hard time keeping it's spatial awareness accurate. Is anyone else struggling with that aspect on this one?

Any recommended sampler preset? because this model is too horny, it takes imitative too fast and without any reason.

If you are still struggling with this. I had really good success by stripping some of the sexual terms in the character cards and prompts. Swapping things out like 'sex scene' with 'NSFW scene', really helped.

Any recommended sampler preset? because this model is too horny, it takes imitative too fast and without any reason.

If you are still struggling with this. I had really good success by stripping some of the sexual terms in the character cards and prompts. Swapping things out like 'sex scene' with 'NSFW scene', really helped.

I'm currently having a great time with stheno v3.2 but I will give lumi a try again once I renew my data plan, thanks!

Any recommended sampler preset? because this model is too horny, it takes imitative too fast and without any reason.

If you are still struggling with this. I had really good success by stripping some of the sexual terms in the character cards and prompts. Swapping things out like 'sex scene' with 'NSFW scene', really helped.

I'm currently having a great time with stheno v3.2 but I will give lumi a try again once I renew my data plan, thanks!

I was going to do the same, try the new Stheno. What presets/prompts are working for you there?

Any recommended sampler preset? because this model is too horny, it takes imitative too fast and without any reason.

If you are still struggling with this. I had really good success by stripping some of the sexual terms in the character cards and prompts. Swapping things out like 'sex scene' with 'NSFW scene', really helped.

I'm currently having a great time with stheno v3.2 but I will give lumi a try again once I renew my data plan, thanks!

I was going to do the same, try the new Stheno. What presets/prompts are working for you there?

The instruct , context and sampler from here are working great: https://www.reddit.com/r/SillyTavernAI/comments/1dff2tl/my_personal_llama3_stheno_presets/

Excellent, those are the ones I collected.

Any way to run it on webUI? I keep geting error

@Zauza25

I'm not much of an Ooba user, but maybe you can try updating the included llama.cpp version there?

Ty, build in updater was saying that all good, so now i reinstall form scratch and it working.

Random Newb Question: Why is there no Q7? Just not part of your normal spread, or not possible due to reasons?

Just not part of the GGUF format spec. It wouldn't likely offer much over the Q6 quants either. Q8 is pretty much indistinguishable from the full model, and Q6 is already very close to that same quality.