Request
I seem unable to make mergekit work very well (or rather, much at all).
In terms of mistral model's the prose of NousResearch/Nous-Capybara-3B-V1.9 and HuggingFaceH4/zephyr-7b-beta are both excellent. Better than anything else I've tried and I've been digging hard into Mistral models. I've made a slerp merge of the two and it's pretty good.
I wondered if you could make a TIES merge of these, or something else than maximizes their shared strengths? Always seemed like TIES was a tad better to me.
Maybe some 120 days, or lima-rp in the mix?
Just a thought....
I seem unable to make mergekit work very well (or rather, much at all).
In terms of mistral model's the prose of NousResearch/Nous-Capybara-3B-V1.9 and HuggingFaceH4/zephyr-7b-beta are both excellent. Better than anything else I've tried and I've been digging hard into Mistral models. I've made a slerp merge of the two and it's pretty good.
I wondered if you could make a TIES merge of these, or something else than maximizes their shared strengths? Always seemed like TIES was a tad better to me.
Maybe some 120 days, or lima-rp in the mix?
Just a thought....
Alright, working on it. but I took Capybara 7B because the 3B isn't mistral my boi.
https://huggingface.co/Undi95/Toppy-M-7B-GGUF Here, you can try this!
Tell me what you think about it
Lol, must have pasted the wrong name for capybara. Thanks so much, I'll definately give it a good try!
Could you merge this Utopia-13B with NousResearch/Nous-Capybara-34B?
I found Utopia-13B is excellent at text generation while NousResearch/Nous-Capybara-34B is good at language understanding.
Maybe their merge will have both of their advantages.
Or could you give me some suggestions on merge(weight/layer), so that I can merge it myself, if you don't have time:)
Could you merge this Utopia-13B with NousResearch/Nous-Capybara-34B?
I found Utopia-13B is excellent at text generation while NousResearch/Nous-Capybara-34B is good at language understanding.
Maybe their merge will have both of their advantages.
Or could you give me some suggestions on merge(weight/layer), so that I can merge it myself, if you don't have time:)
We sadly can't merge 34B and 13B, it's impossible.