turboderp's picture
Upload 17 files
dd0cc25 verified
|
raw
history blame
838 Bytes
# Turbcat 8b
# Release notes
This is a direct upgrade over cat 70B, with 2x the dataset size, added Chinese support.
# Data Generation
The data generation process is largely the same. Except additional Chinese data were added. 20 postdocs participated in the annotation process and standard model training for embedding scoring.
# Task coverage
In addition to standard assistant and roleplay data, the following tasks are targeted:
* GRE
* SAT
* MCAT
* Chinese Kaoyan
# Thirdparty dataset
Thanks to the following people for their tremendous support for dataset generation:
* steelskull for the medical COT dataset with gpt4o
* Gryphe for the wonderful action packed dataset
* Turbca for being turbca
# Prompt format for 8b:
llama3
# Prompt format for 72b:
chatml
# Support
Please join https://discord.gg/DwGz54Mz for model support