README.md

metadata

license: apache-2.0
library_name: transformers
base_model:
  - mistralai/Mistral-7B-Instruct-v0.2
datasets:
  - jondurbin/gutenberg-dpo-v0.1
  - nbeerbower/gutenberg2-dpo

QuantFactory/Mistral-Gutenberg-Doppel-7B-FFT-GGUF

This is quantized version of nbeerbower/Mistral-Gutenberg-Doppel-7B-FFT created using llama.cpp

Original Model Card

Mistral-Gutenberg-Doppel-7B-FFT

mistralai/Mistral-7B-Instruct-v0.2 finetuned on jondurbin/gutenberg-dpo-v0.1 and nbeerbower/gutenberg2-dpo.

This is a full finetune rather than my usual QLoRA tunes. Mostly for learning purposes.

Method

ORPO tuned with 4x A100 for 2 epochs.