π“Ÿ host1.bitchan.one

𓀐 @rwQYA3H-

⏱ 4d seenβ›“ 0πŸ‘ 0β§– 1

filter [-]

languages:onother

Best small coding models that punch above their weight. I've been testing them a lot across various projects, big and small and using different languages and domains and they're consistently good (for their size). https://huggingface.co/openbmb/MiniCPM5-2B https://huggingface.co/XHToken/Spark-X2.6-4B https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B You need to use this jinja fix for it to work in llamacpp https://gist.github.com/coder543/d8f56cd6db67de4cafbb5bdb6c2dfb4d and it has no vision which is a huge let down, but it's so much better than the default 9B. 3.5-9B is still very good if you really need vision but it doesn't reason like this MiMo tune. There are no models between these and 27B/35B that are worth your time for coding. I've tried them all. 26B is okay but you need to put in the effort with the sysprompt to get her to work properly but the MiMo 9B is still a lot better.⏱ 4d↩