Model Overview

MaziyarPanahi/calme-2.7-qwen2-7b is a 7.6 billion parameter language model, fine-tuned by MaziyarPanahi from the Qwen/Qwen2-7B base model. The primary objective of this fine-tuning was to achieve improved performance across a range of benchmarks, making it a more robust and capable general-purpose model.

Key Capabilities & Performance

This model demonstrates enhanced capabilities as indicated by its evaluation results on the Open LLM Leaderboard. Notable scores include:

• Avg. Score: 22.07
• IFEval (0-Shot): 35.92
• BBH (3-Shot): 28.91
• MMLU-PRO (5-shot): 30.06

These metrics suggest improved reasoning and instruction following compared to the base model. The model supports a substantial context length of 131072 tokens, allowing for processing longer inputs and generating more extensive outputs.

Usage and Prompt Format

The model utilizes the ChatML prompt template, which is a common and effective format for instruction-tuned models. This template structures conversations with explicit roles for system, user, and assistant messages, facilitating clear interaction and response generation.

For users interested in quantized versions, GGUF models are available at MaziyarPanahi/calme-2.7-qwen2-7b-GGUF.