| Property | Value |
|---|---|
| License | Other |
| Language | English |
| Framework | PyTorch |
| Dataset | WizardLM Evol Instruct V2 |
What is WizardLM-33B-V1.0-Uncensored?
WizardLM-33B-V1.0-Uncensored is an advanced language model that represents a retraining of the original WizardLM-30B-V1.0 with a filtered dataset. The model is specifically designed to reduce refusals, avoidance, and bias while maintaining high performance across various benchmarks. It utilizes the Vicuna-1.1 style prompts for interaction and achieves impressive scores on multiple evaluation metrics.
Implementation Details
The model is built on the LLaMA architecture and implements text generation using PyTorch. It's trained using the WizardLM Evol Instruct V2 dataset and employs a specific prompt format for interaction.
- Achieves 59.99% average performance across major benchmarks
- Implements Vicuna-1.1 style prompting system
- Built with text-generation-inference capabilities
- Optimized for reduced content filtering while maintaining performance
Core Capabilities
- Strong performance on HellaSwag (83.84% accuracy)
- Robust reasoning capabilities with ARC Challenge (63.65%)
- Excellent performance on Winogrande (77.66% accuracy)
- Competitive MMLU performance (59.36% accuracy)
- Reliable truthfulness with TruthfulQA (56.8%)
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its uncensored nature while maintaining high performance across benchmarks. It's specifically designed to reduce typical AI model limitations like refusals and excessive filtering, while still preserving core capabilities.
Q: What are the recommended use cases?
The model is suitable for various text generation tasks, particularly where unrestricted outputs are desired. However, users should note that they are responsible for the content generated, and the model should be used with appropriate consideration for ethical implications.