File size: 1,323 Bytes
1be535b a945cdd 1be535b a945cdd 1be535b 0c9da9f a945cdd 1be535b a945cdd 1be535b a945cdd 1be535b a945cdd 1be535b a945cdd 1be535b a945cdd 1be535b 35cf69f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 |
---
library_name: transformers
license: apache-2.0
base_model:
- nbeerbower/Mahou-1.5-mistral-nemo-12B-lorablated
datasets:
- nbeerbower/Arkhaios-DPO
- nbeerbower/Purpura-DPO
---
> 🧪 **Just Another Model Experiment**
>
> This is one of many experimental iterations I'm sharing publicly while I mess around with training parameters and ideas. It's not a "real" release - just me being transparent about my learning process. Feel free to look under the hood, but don't expect anything production-ready!
# Mistral-Nemo-Prism-12B-v7
[Mahou-1.5-mistral-nemo-12B-lorablated](https://huggingface.co/nbeerbower/Mahou-1.5-mistral-nemo-12B-lorablated) finetuned on [Arkhaios-DPO](https://huggingface.co/datasets/nbeerbower/Arkhaios-DPO) and [Purpura-DPO](https://huggingface.co/datasets/nbeerbower/Purpura-DPO).
The goal was to reduce archaic language and purple prose in a completely uncensored model.
### Method
[ORPO tuned](https://mlabonne.github.io/blog/posts/2024-04-19_Fine_tune_Llama_3_with_ORPO.html) with 8x A40 for 10 epochs.
For this version, beta was increased to 2.
**In conclusion, LoRA does not seem to be able to completely remove some of the language issues deeply embedded in the model.** |