garage-bAInd/Camel-Platypus2-13B
Camel-Platypus2-13B is a 13 billion parameter auto-regressive language model based on the LLaMA 2 transformer architecture, created by merging garage-bAInd/Platypus2-13B and augtoma/qCammel-13. Instruction fine-tuned using LoRA, it leverages a STEM and logic-based dataset for enhanced performance. This model is designed for general English language tasks, with a focus on reasoning and factual recall, and has a context length of 4096 tokens.
Loading preview...
Model Overview
Camel-Platypus2-13B is a 13 billion parameter instruction-tuned language model built upon the LLaMA 2 transformer architecture. It is a merge of two distinct models: garage-bAInd/Platypus2-13B and augtoma/qCammel-13. The model was fine-tuned using LoRA on a single A100 80GB GPU, leveraging the garage-bAInd/Open-Platypus dataset, which is known for its STEM and logic-based content.
Key Capabilities
- Reasoning and Logic: Benefits from training on STEM and logic-focused datasets, suggesting proficiency in these areas.
- Instruction Following: Instruction fine-tuned to respond effectively to user prompts.
- General Language Tasks: Capable of handling a wide range of English language generation and understanding tasks.
Performance Highlights
Evaluated on the Open LLM Leaderboard, Camel-Platypus2-13B achieved an average score of 52.12. Notable scores include:
- ARC (25-shot): 60.75
- HellaSwag (10-shot): 83.61
- MMLU (5-shot): 56.51
- TruthfulQA (0-shot): 49.6
When to Use This Model
This model is suitable for applications requiring a 13B parameter model with a strong foundation in reasoning and instruction following, particularly for tasks that benefit from exposure to scientific and logical data. Its LLaMA 2 base and specific fine-tuning make it a candidate for general-purpose English language generation where factual accuracy and logical coherence are important. Developers should perform safety testing tailored to their specific applications due to the inherent risks associated with LLMs.