One of the best LLMs out there !💥
This LLM is really incredible. Well balanced, not too verbose, nor too shy or uninterested. It's sober, without crazy approaches, it's not psychotic like some which talk like drunk people, but it provides a decent, real, open conversation about every subject, keeping the interest, without repeating itself or running out of ideas. 🙏👍💥
agreed. I've been dabbling with running local models and this one is one of my favorites by far.
Really glad to hear! =) This is the kind of response that makes me thinking about release a new and improved model! Cheers
That would be awesome. ❤️ It's sad to see so many bad LLMs being updated on HG every day, so frequently, while some good ones are kept forgotten. Perhaps this model could be an excellent base model to mix with some other good ones, to create an "omelette" of good reasoning, intelligent and well balanced uncensored LLM. There are already too many ero oriented LLMs like Umbral, or Omelette, they are very good on RP, but we always need other general LLM for everything else, with excellent reasoning, intelligence, education, interested, non repetitive, and with great creativity. 🙏💥👍❤️
Agreed. You did a phenomenal job, haven't found a more compliant model.
Hello, you guys do you have any Advice other uncensored AI?
Excelente, espero con ansias la actualización a 3.1 :D
@DarkClaw42 The 3.1 has been released already :) https://huggingface.co/Orenguteng/Llama-3.1-8B-Lexi-Uncensored-GGUF
Really new to this and maybe it's because I'm using Anything LLM but at least, when dealing with me, it has a different interpretation of what "compliant" means. I can sometimes work around it by asking it to roleplay, but that creates other issues (like suddenly switching to a character from earlier and flat out ignoring the prompt).
There is also the issue that it seems to forget settings and characters it has been introduced to. It will work for a while, but then suddenly people are no longer on a spaceship but in a fancy office place in some random city. Some characters stay consistent, but others are randomly replaced. And so far I haven't been able to really get it to adopt a certain language pattern. It will scratch the surface at times, but that's about it.
I've only downloaded the large (16GB) file. Do I need additional files? Or is it just an insane amount of trial & error?
Really new to this and maybe it's because I'm using Anything LLM but at least, when dealing with me, it has a different interpretation of what "compliant" means. I can sometimes work around it by asking it to roleplay, but that creates other issues (like suddenly switching to a character from earlier and flat out ignoring the prompt).
There is also the issue that it seems to forget settings and characters it has been introduced to. It will work for a while, but then suddenly people are no longer on a spaceship but in a fancy office place in some random city. Some characters stay consistent, but others are randomly replaced. And so far I haven't been able to really get it to adopt a certain language pattern. It will scratch the surface at times, but that's about it.I've only downloaded the large (16GB) file. Do I need additional files? Or is it just an insane amount of trial & error?
Oh yes, that's because the llama-3 8B by META is limited to 8K context length only. Meaning it will forget or become incoherent after 8K length of conversations. Some clients (I think such as LM Studio also) have rope scaling that can extend the context , you should be able to double that with somehow good quality without losing much coherency meaning you can run longer convos without the model forgetting the history (context)
Otherwise, I would recommend you try out the Llama 3.1 V2 which I also uploaded, it has 130K context length meaning a MASSIVE upgrade to the context length. However, it's a new model in comparison to this one, so the "personality" might differ.
Edit:
By the way, people here have been complaining about Anything LLM and settings things up with that , try running a different client, I would recommend LM Studio for new beginners, i use it even myself sometimes.
Really new to this and maybe it's because I'm using Anything LLM but at least, when dealing with me, it has a different interpretation of what "compliant" means. I can sometimes work around it by asking it to roleplay, but that creates other issues (like suddenly switching to a character from earlier and flat out ignoring the prompt).
There is also the issue that it seems to forget settings and characters it has been introduced to. It will work for a while, but then suddenly people are no longer on a spaceship but in a fancy office place in some random city. Some characters stay consistent, but others are randomly replaced. And so far I haven't been able to really get it to adopt a certain language pattern. It will scratch the surface at times, but that's about it.I've only downloaded the large (16GB) file. Do I need additional files? Or is it just an insane amount of trial & error?
Oh yes, that's because the llama-3 8B by META is limited to 8K context length only. Meaning it will forget or become incoherent after 8K length of conversations. Some clients (I think such as LM Studio also) have rope scaling that can extend the context , you should be able to double that with somehow good quality without losing much coherency meaning you can run longer convos without the model forgetting the history (context)
Otherwise, I would recommend you try out the Llama 3.1 V2 which I also uploaded, it has 130K context length meaning a MASSIVE upgrade to the context length. However, it's a new model in comparison to this one, so the "personality" might differ.
Edit:
By the way, people here have been complaining about Anything LLM and settings things up with that , try running a different client, I would recommend LM Studio for new beginners, i use it even myself sometimes.
Wow. thanks for the quick reply. I guess I'll try 3.1then. And I'll make sure to have a look at LM Studio. Hopefully I can make sure, spaceships become office spaces when a specific sci-fi plot, involving a multiverse, requires them to do so...
I recently came across Orenguteng/Llama-3-8B-Lexi-Uncensored-GGUF and am very impressed by its potential. I am eager to test its utility using some widely recognized benchmarks, specifically:
GPQA for general knowledge evaluation
MMLU for academic knowledge assessment
IFEval for instruction-following abilities
Could you please provide any guidance on it. Practically how can i measure them on the model?