openerotica/

Llama-3.3-70B-Instruct-multiturn-exl2-2.25bpw

Quantized with these options:

python convert.py
-i /workspace/text-generation-webui/models/meta-llama_Llama-3.3-70B-Instruct
-o /workspace/temp/
-m /workspace/measurement.json
-l 16000
-cf /workspace/2.25
-b 2.25
-c /workspace/converted_rp.parquet
-r 100

Dataset is long context synthetic roleplay multiturn dataset pre formatted with llama instruct role added between each turn.

https://huggingface.co/datasets/openerotica/multi-turn-aware-quantization-llama-3.3-rp-test