This is the GOAT in this param range.
Great model. Beats all the llama 2 13b models I tried. I recently switched to fimbulvetr-kuro-lotus, but this model is still fantastic.
Most consistent model I've ever used. The only downside is that 4K context, but you can scale it to 6144 context and only lose a little perplexity/coherence.
It does well with following instructions, character cards, and general understanding. Punches way above its weight. Can do NSFW but doesn't force you into it. Horny when it should be.
This and the
In certain ways, it feels smarter than any 13B I’ve used. I always thought that <13B will always be worse, but was surprised with this one, and it runs faster as a bonus for being 11B. It writes very well, has some of the most strong, coherent, interesting and fun output I’ve seen, and understands things most 13Bs didn’t. I have to note I used Fimbulvetr-11B-v2, not v1.
This is absolutely garbage. The only thing that its good for is maybe basic conversation. Otherwise, I don't recommend this model.
Punches way above its weight, and demonstrates incredible coherence, reasoning, and even can keep track of several characters at once.
Only one that seems to have common sense and spacial awareness. I think all solar models have a lot of common sense. Whatever they did works, and they need to continue researching it. This small model feels like it's a 100b are at least something 10x its size.