posicube/Llama2-chat-AYT-13B: An Ensembled Llama-2 Chat Model

This model, developed by Posicube Inc., is a 13 billion parameter variant based on the Llama-2-13b-chat-hf architecture. Its core innovation lies in its ensemble approach, where it integrates the strengths of top-performing models across key benchmarks such as ARC, MMLU, and TruthfulQA. This strategy aims to maximize overall performance by combining diverse capabilities.

Key Capabilities & Performance

• Ensemble Fine-tuning: Leverages an ensemble method to integrate top-ranked models from various benchmarks.
• Strong Benchmark Scores: Achieved top-ranker status among 13B models on September 13th, 2023, with competitive scores:
  • ARC (25-shot): 63.57
  • HellaSwag (10-shot): 83.77
  • MMLU (5-shot): 59.69
  • TruthfulQA (0-shot): 55.48
• Dataset Training: Fine-tuned using Orca-style and Alpaca-style datasets.

Use Cases & Considerations

This model is well-suited for general conversational AI, question-answering, and tasks requiring robust performance across multiple reasoning and knowledge domains. As with all Llama 2 variants, it carries inherent limitations and biases. Developers should conduct thorough safety testing and tuning for specific applications, as potential outputs cannot be fully predicted. The model is bound by the original Llama-2 license and comes without warranty.