license: apache-2.0
tags:
- merge
- roleplay
- exl2
- not-for-all-audiences
RP-Stew-v4.0-34B
Base model:
https://huggingface.co/ParasiticRogue/RP-Stew-v4.0-34B
Parquet used (Bluemoon-Light/Chat-Vicuna-1.1) for quantization:
https://huggingface.co/datasets/ParasiticRogue/Bluemoon-Light
Another experimental/testing merge and quant to try and increase Stew's capabilities, but with some slight alterations in models used, and this one actually seems to show a bit more promise than v3 with the brief tests done so far.
trust-remote-code must be turned on for this version still due to the base model being Capybara, but I'll look into fixing this later if it performs comparably to v2 or better during further testing.
Universal Light's settings of Temperature at 1.25/Min-P 0.1 (decrease Temp and increase Min-P by about .1~ if it start hallucinating too much) with Smoothing Factor at 0.3 and Smoothing Curve at 1.5 plus DRY multiplier at 0.8 seems stable for now.
Prompt Format: Chat-Vicuna-1.1
SYSTEM: {system_prompt}<|end|>
USER: {prompt}<|end|>
ASSISTANT: {output}<|end|>
Models Merged
The following models were included in the merge:
https://huggingface.co/NousResearch/Nous-Capybara-34B
https://huggingface.co/migtissera/Tess-2.0-Yi-34B-200K
https://huggingface.co/jondurbin/bagel-dpo-34b-v0.5
https://huggingface.co/maywell/PiVoT-SUS-RP
https://huggingface.co/Sao10K/NyakuraV2-34B-Yi-Llama
https://huggingface.co/NeverSleep/CausalLM-RP-34B
https://huggingface.co/adamo1139/Yi-34b-200K-AEZAKMI-RAW-TOXIC-2702