main : 8bpw h8
6bh8 : 6bpw h8
4bh8 : 4bpw h8

Using VMware/open-instruct as dataset

Quantization settings : python convert.py -i models/maywell_PiVoT-0.1-early -o PiVoT-0.1-early-temp -cf PiVoT-0.1-early-8bpw-h8-exl2 -c 0000.parquet -l 4096 -b 8 -hb 8
python convert.py -i models/maywell_PiVoT-0.1-early -o PiVoT-0.1-early-temp2 -cf PiVoT-0.1-early-6bpw-h8-exl22 -c 0000.parquet -l 4096 -b 6 -hb 8 -m PiVoT-0.1-early-temp/measurement.json
python convert.py -i models/maywell_PiVoT-0.1-early -o PiVoT-0.1-early-temp3 -cf PiVoT-0.1-early-4bpw-h8-exl2 -c 0000.parquet -l 4096 -b 4 -hb 8 -m PiVoT-0.1-early-temp/measurement.json

PivoT is Finetuned model based on Mistral 7B. It is variation from Synatra v0.3 RP which has shown decent performance.

OpenOrca Dataset used when finetune PiVoT variation. Arcalive Ai Chat Chan log 7k, ko_wikidata_QA, kyujinpy/OpenOrca-KO and other datasets used on base model.

Follow me on twitter: https://twitter.com/stablefluffy

Consider Support me making these model alone: https://www.buymeacoffee.com/mwell or with Runpod Credit Gift 💕

Contact me on Telegram: https://t.me/AlzarTakkarsen