Tried using DS-R1-Distill-70B-ArliAI-RpR-v4-Large-exl2 4.25BPW over the night for 2x4090 for RPing.

Result: It's worse than Mixtral-8x7B-Instruct-v0.1-LimaRP-ZLoss-6.0bpw-h6-exl2-rpcal from like 2 years ago. Wasted so much tokens in the <think> portion and then output less coherent responses.

I guess 8x7B mixtral is still my go to for RPing. It's over...