metadata
license: apache-2.0
datasets:
- Setiaku/Stheno-v3.2
- Squish42/bluemoon-fandom-1-1-rp-cleaned
- openerotica/freedom-rp
- MinervaAI/Aesir-Preview
- jeiku/JeikuL3v2
- ResplendentAI/Sissification_Hypno_1k
- ResplendentAI/Synthetic_Soul_1k
- ResplendentAI/theory_of_mind_fixed_output
language:
- en
tags:
- not-for-all-audiences
Nymph 8B
GGUF here: https://huggingface.co/mradermacher/Nymph_8B-i1-GGUF
- Mobile (ARM): Q4_0_X_X
Nymph is the culmination of everything I have learned with the T-series project. This model aims to be a unique and full featured RP juggernaut.
The finetune incorporates 1.6 Million tokens of RP data sourced from Bluemoon, FreedomRP, Aesir-Preview, and Claude Opus logs. I made sure to use the multi-turn sharegpt datasets this time instead of alpaca conversions. I have also included three of my personal datasets. The final touch is an ORPO based upon Openhermes Roleplay preferences.