Mistral CEO confirms 'leak' of new open source AI model nearing GPT4 performance

Re:Mistral CEO confirms 'leak' of new open source AI model n... by MallocVoidstar on Wednesday January 31, 2024 @03:59PM (#39209336)

That isn't unquantized, it's de-quantized. They went from Q5 to fp16 for use in Pytorch instead of the GGUF ecosystem.

[ Parent | on HN ]

Re:Mistral CEO confirms 'leak' of new open source AI model n... by Taek on Wednesday January 31, 2024 @04:03PM (#39209397)

I never thought people would be upscaling models by increasing quantization precision. The rationale makes sense bit its also a goofy outcome.

[ Parent | on HN ]

Re:Mistral CEO confirms 'leak' of new open source AI model n... by nullc on Wednesday January 31, 2024 @08:09PM (#39211749)

You should be able to upscale and fine tune to recover performance, I suppose!

Clearly we should train a diffusion model to denoise the weights of LLM transformer models. Yo dawg.

[ Parent | on HN ]

Re:Mistral CEO confirms 'leak' of new open source AI model n... by throwaway9274 on Wednesday January 31, 2024 @04:46PM (#39209853)

Yes, that’s correct. Good correction.

[ Parent | on HN ]