Model Overview

dhmeltzer/Llama-2-13b-hf-eli5-wiki-1024_r_64_alpha_16_merged is a 13 billion parameter model built upon the Llama-2 architecture, designed for general-purpose language tasks. It incorporates a 4096-token context window, allowing it to process and generate moderately long sequences of text.

Key Capabilities & Performance

This model has been evaluated on the Open LLM Leaderboard, showcasing its performance across several benchmarks. Its average score is 46.93.

• ARC (25-shot): Achieves 58.96, indicating proficiency in common sense reasoning.
• HellaSwag (10-shot): Scores 81.94, demonstrating strong performance in natural language inference.
• MMLU (5-shot): Reaches 55.0, reflecting its ability in multi-task language understanding.
• TruthfulQA (0-shot): Scores 40.26, assessing its capacity for truthful responses.
• Winogrande (5-shot): Achieves 76.56, showing competence in resolving pronoun ambiguity.
• GSM8K (5-shot): Scores 8.72, suggesting limited capability in mathematical reasoning.
• DROP (3-shot): Scores 7.05, indicating areas for improvement in reading comprehension with discrete answers.

Use Cases

Given its balanced performance across various benchmarks, this model is suitable for:

• General text generation and understanding tasks.
• Applications requiring common sense reasoning and natural language inference.
• Educational tools for explaining concepts (ELI5 - Explain Like I'm 5) and summarizing information from Wikipedia-like sources, as suggested by its fine-tuning focus.