airoboros-13b
| Property | Value |
|---|---|
| Base Model | LLaMa 13B |
| License | CC-BY-NC-4.0 (Research Only) |
| Evaluation Score | 98.087 (GPT-3.5 adjusted) |
| Primary Use | Research/Non-Commercial |
What is airoboros-13b?
airoboros-13b is an experimental language model built on the LLaMa 13B architecture, fine-tuned using synthetic training data generated through a novel approach. The model achieves impressive performance, scoring 98.087 on GPT-3.5 adjusted metrics, demonstrating its capability to handle a wide range of tasks.
Implementation Details
The model utilizes a unique training approach where synthetic data was generated using a "jailbreak" prompt methodology. This approach successfully expanded the range of topics and reduced response refusals compared to traditional training methods. The model follows the FastChat/vicuna prompt format and can be used with or without system prompts.
- Compatible with FastChat/vicuna prompt format
- Trained on synthetically generated data
- Achieves near GPT-3.5 level performance
- Research-focused implementation
Core Capabilities
- Broad topic coverage and reduced refusal rates
- High evaluation scores across diverse prompts
- Flexible prompt formatting support
- Detailed and comprehensive responses
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its training methodology using synthetic data generated through a specialized prompt approach, resulting in broader response capabilities while maintaining high-quality outputs.
Q: What are the recommended use cases?
The model is strictly intended for research purposes only, as specified by its license. It cannot be used for commercial applications due to both LLaMa's research license restrictions and OpenAI's data usage terms.
