e88 88e d8
d888 888b 8888 8888 ,"Y88b 888 8e d88
C8888 8888D 8888 8888 "8" 888 888 88b d88888
Y888 888P Y888 888P ,ee 888 888 888 888
"88 88" "88 88" "88 888 888 888 888
b
8b,
e88'Y88 d8 888
d888 'Y ,"Y88b 888,8, d88 ,e e, 888
C8888 "8" 888 888 " d88888 d88 88b 888
Y888 ,d ,ee 888 888 888 888 , 888
"88,d88 "88 888 888 888 "YeeP" 888
PROUDLY PRESENTS
Llama-3-8B-Instruct-DADA-exl2-rpcal
Quantized using 200 samples of 8192 tokens from an RP-oriented PIPPA dataset.
Branches:
main
--measurement.json
8b8h
-- 8bpw, 8bit lm_head6b6h
-- 6bpw, 6bit lm_head4b6h
-- 4bpw, 6bit lm_head
Original model link: Envoid/Llama-3-8B-Instruct-DADA
Original model README below.
Llama-3-8B-Instruct-DADA
Warning: This model is experimental and thus potentially unpredictable.
This model employs the same strategy as Mixtral Instruct ITR DADA
I trained Llama-3-8B-Instruct on the Alpaca-DADA dataset for 10 epochs at 1e-6 learning rate. I then did a 50/50 SLERP merge of the resulting model back onto Llama-3-8B-Instruct
This model may require custom stopping strings to tame due to current issues surrounding Llama-3 EOS tokens and various back-ends. It certainly gives some interesting answers using an assistant template/card in SillyTavern, though.
The below answer is one of the more interesting answers I've gotten out of an LLM on the same query, although there was an indentiation error (indicated by the red circle)
Training was done using qlora-pipe