pe-nlp/llama-2-13b-platypus-vicuna-wizard
The pe-nlp/llama-2-13b-platypus-vicuna-wizard is a 13 billion parameter language model based on the Llama 2 architecture, featuring a 4096-token context length. This model is a merge of Platypus, Vicuna, and WizardLM, designed to combine their respective strengths in instruction following, conversational abilities, and complex reasoning. It demonstrates competitive performance across various benchmarks, making it suitable for general-purpose conversational AI and instruction-tuned applications.
Loading preview...
Model Overview
The pe-nlp/llama-2-13b-platypus-vicuna-wizard is a 13 billion parameter language model built upon the Llama 2 architecture, offering a 4096-token context window. This model represents a strategic merge of three distinct instruction-tuned models: Platypus, Vicuna, and WizardLM. The intention behind this merge is to leverage the individual strengths of each component, aiming for a more robust and versatile model capable of handling a broader range of tasks.
Key Capabilities & Performance
Evaluated on the Open LLM Leaderboard, this model exhibits a balanced performance across various benchmarks, indicating its general utility. Notable scores include:
- ARC (25-shot): 61.26
- HellaSwag (10-shot): 82.31
- MMLU (5-shot): 55.21
- TruthfulQA (0-shot): 41.91
- Winogrande (5-shot): 75.77
While its performance on specific reasoning tasks like GSM8K (0.91) and DROP (44.96) suggests areas for further specialization, its overall average score of 51.76 positions it as a capable model for diverse applications.
Ideal Use Cases
This model is well-suited for scenarios requiring:
- General-purpose instruction following: Benefiting from the combined instruction-tuning of its merged components.
- Conversational AI: Leveraging Vicuna's strengths in dialogue generation.
- Complex reasoning tasks: Drawing on WizardLM's capabilities for more intricate problem-solving, though specific benchmark scores should be considered for highly specialized applications.