4.65bpw quantization of Euryale 1.3-L2-70B (https://huggingface.co/Sao10K/Euryale-1.3-L2-70B), using exllamav2.
Original model card
17th Attempt. Past 10 Failed, cost me >$200 lol.
Idea is an updated version of Euryale with ReMantik instead of the ties-merge between the original 3 models.
This is then mixed with a saucy model with a Mythomax-esque Ratio, and a certain experimental LoRA applied to it.
Test Results: Works Well.
NSFL and NSFW fine in roleplay context.
slight censor with 0 context, zero issues in actual RP / ERP.
Good Prose, Not Dumbed Down due to RP merges from testing.
I have not encountered any repetition issues some had with the original Euryale. tell me if you do, though.
Prompt and System Format: most works well. I recommend Alpaca. ST Settings used for Test: Lightning 1.1 System Prompt + Shortwave(1.2 Temperature)
Support the original author here :)
- Downloads last month
- 7