w4r10ck/SOLAR-10.7B-Instruct-v1.0-uncensored
w4r10ck/SOLAR-10.7B-Instruct-v1.0-uncensored is a 10.7 billion parameter instruction-tuned causal language model, derived from the SOLAR-10.7B-Instruct-v1.0 architecture. This model has been fine-tuned using Lora and DPOTrainer to reduce censorship, making it suitable for applications requiring less restrictive content generation. It maintains a context length of 4096 tokens and is optimized for general instruction-following tasks with a focus on uncensored responses.
Loading preview...
Model Overview
w4r10ck/SOLAR-10.7B-Instruct-v1.0-uncensored is a 10.7 billion parameter instruction-tuned model based on the original upstage/SOLAR-10.7B-Instruct-v1.0. The primary differentiation of this version is its fine-tuning to be less censored, achieved through training with Lora and DPOTrainer on the unalignment/toxic-dpo-v0.1 dataset.
Key Characteristics
- Base Model: Derived from SOLAR-10.7B-Instruct-v1.0.
- Parameter Count: 10.7 billion parameters.
- Context Length: Supports a context window of 4096 tokens.
- Fine-tuning: Utilizes Lora and DPOTrainer for instruction-following and reduced censorship.
- Training Data: Fine-tuned on
unalignment/toxic-dpo-v0.1to achieve its uncensored nature.
Performance Insights
Evaluations on the Open LLM Leaderboard indicate an average score of 20.56. Specific metric scores include:
- IFEval (0-Shot): 38.84
- BBH (3-Shot): 33.86
- MMLU-PRO (5-shot): 26.04
Use Cases
This model is particularly suited for applications where a less restrictive and more direct response generation is desired, especially in scenarios where the base model's inherent censorship might be a limiting factor. Developers can refer to the original upstage/SOLAR-10.7B-Instruct-v1.0 for general usage instructions and further model details.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.