Single Bit LLMs

Single-bit precision autoregressive somewhat large language models 
for research purposes maybe this is something we will be using in the near future and future generations will go why the fuck were these people using 16 bits of precision
---
https://arxiv.org/abs/2402.17764
https://arxiv.org/pdf/2310.11453.pdf
---
There are some graphs toward the starting of the paper, I would want to verify those so here it is a reproduction with mistral 7B V0.1

Turns out graphing MMLU or any EVAL for that matter for quantizer models is pretty slow, either I wiill make this fast now or maybe after I'm done with the paper 

anyways here are all the quantized models - https://huggingface.co/shauray/mistral_GGUF_TEST01
to be done 

---
link to the blog 
---