metadata
license: apache-2.0
base_model:
- JungZoona/T3Q-qwen2.5-14b-v1.0-e3
Decensored using a custom training script guided by activations, similar to ablation/"abliteration" scripts but not exactly the same approach.
Original model is #14 ranked on Open LLM Leaderboard
The training script is released under the MIT license: https://github.com/nkpz/DeLMAT - This model was built using an upcoming development version
This is overtrained toward acceptance, and it overuses the words like "absolutely" and "sure". I don't think I'd recommend it for creative writing for this reason, but it's decent for question answering and getting it to say bizarre things. I almost decided not to release this because it's so unbalanced, but I think there's something kind of fun and interesting about always getting an affirmative response.