Reminds me of https://github.com/AlexsJones/llmfit

does this work similar to airllm? i am wondering how it would handle something like quantizing kimi k3 on a budget of 8 gbs, or is that something you are not attempting to solve yet?

tried it out but based on the model sizing result i got i got an insufficient memory error when the server started running