This is quantized version of ezelikman/quietstar-8-ahead created using llama.cpp

Mistral-7b with continued pretraining using Quiet-STaR (https://arxiv.org/abs/2403.09629) for generating 8 thought tokens before each output token.