| Property | Value |
|---|---|
| Base Architecture | Llama 2 |
| Parameters | 13B |
| License | CC BY-NC 4.0 |
| Language | English |
| Paper | Platypus Paper |
What is OpenOrca-Platypus2-13B?
OpenOrca-Platypus2-13B represents a groundbreaking merger between garage-bAInd/Platypus2-13B and Open-Orca/OpenOrcaxOpenChat-Preview2-13B, creating a powerful language model that excels in both STEM reasoning and general knowledge tasks. The model achieves impressive benchmark scores, including 59.5 on MMLU and 62.88 on ARC, positioning it as a leading performer in its category.
Implementation Details
Built on the Llama 2 architecture, this model combines specialized training from two distinct datasets: the STEM-focused Open-Platypus dataset and the refined GPT-4 data from OpenOrca. The implementation supports various prompting templates and can be deployed using the Transformers library.
- Instruction fine-tuned using LoRA on A100-80GB GPU
- Supports both Platypus and OpenChat prompt formats
- Optimized for both academic and general-purpose tasks
Core Capabilities
- Strong performance in STEM and logic-based reasoning
- Enhanced truthfulness with 52.69% on TruthfulQA
- Robust performance on AGIEval (112% of base model)
- Superior LSAT Logical Reasoning capabilities
- Impressive general knowledge demonstrated by MMLU scores
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its combination of STEM expertise from Platypus with the broad knowledge base of OpenOrca, resulting in superior performance across diverse tasks while maintaining high accuracy in specialized domains.
Q: What are the recommended use cases?
The model excels in academic and research applications, particularly in STEM fields, logical reasoning tasks, and general knowledge queries. It's particularly well-suited for educational applications and complex problem-solving scenarios.