Congratulations. The best one so far ! 💥
I test some of the most important GGUF models people begin to talk about. Most from mradermacher, bartowski, etc. Recently, I'm testing L3 Stheno v3... L3 umbral mind RP v3... They are good, but 2 models blew my mind, called: mradermacher Llama 3 Lexi Smaug Uncensored ( excellent ), and mradermacher uncen merger omelette RP v0 1 ( it blew out of the water in knowledge, RP, it's the most well balanced model I tested on LM Studio so far among toooooo many. A real state of the art model, much better than L3 umbral mind RP in all ways ).
I can't wait to test this new updated v 0.2 version, when someone create the GGUF version, since it's the one which runs on LM Studio.
Some black magic is at work here. I was very skeptical at first, but this model is now actually one of my favorites, with great spatial awareness and prompt adherence.
Yes, and have you notice that like Umbral, it also brings new elements underlying the plot that wasn't initially aware even by yourself ? It's very different from other LLMs which keeps always inside the scenario you proposed. This one and Umbral extrapolates the plot not to change it, but to bring underlying conditions and activities that are running in parallel, that we often weren't aware of. This is intelligence. 🙏👍❤️
Some black magic is at work here.
That would be my gut. I tend to rely more on my gut rather than thinking too much about how to go about merging something as I think it would make me make too many safe decisions.
I'm always aiming for something fun and creative rather than a highly intelligent RP LLM as I've always had the problem where after using so many models, everything started blurring together and almost all of them started to feel the same to me, especially mergers. It's because of this that I tend to go schizo with the number of models and merging steps, blending so many things together makes the model feels like it's its own thing rather than just some merger of a few models.
I will say though, that even though my goal was creativity and new writing styles, the last thing I was expecting to get was decent intelligence as well. I went into this thinking I would have to do more heavy lifting in RPs.
Some black magic is at work here.
That would be my gut. I tend to rely more on my gut rather than thinking too much about how to go about merging something as I think it would make me make too many safe decisions.
Congratulations ! I don't know how you did that, but sometimes, serendipity happens. I had the pleasure to test some of your models in GGUF ( LM Studio ), which behave quite different from all the rest. It's like you said, after some time, most models look alike, but yours are different. I don't know how to describe them, but they seem to have a proper life, kind of unexpected intelligence. Perhaps, either you stumbled on something really new that nobody else is perceiving, or the result was by chance. These models should be studied, in order that this characteristic or technique could be improved even more. Many times in history, some people had the luck to discover something new just by experimentation. And you have talent for this thing, like few people I have seen on the community. Your models are a "full house" here. Perhaps more people will discover them, I hope. We need more YouTube reviews about models. People only see Dolphin... But it can't handle a candle to your models, you know that. 🙏👍❤️
