LoneStriker's picture
ExLLaMA V2 quant of Yi-34B-GiftedConvo-merged-6.0bpw-h6-exl2
4356f7c
metadata
license: mit
datasets:
  - NobodyExistsOnTheInternet/GiftedConvoBeforeEcons

Trained on over 20k instruct generated all by gpt-4 or humans

Dataset features: 1000 long evolved conversations based off LIMA Subsection of correct PRM800k data Subsection of CamelAI's Physics and Chemistry data

The model is trained with Qlora as well as Axolotl.

The model format is Vicuna 1.1:

User: ...
Assistant: ...