Model Overview

Weyaxi/OpenHermes-2.5-neural-chat-7b-v3-2-7B is a 7 billion parameter language model developed by Weyaxi. It is a product of a TIES merge between two prominent models: teknium/OpenHermes-2.5-Mistral-7B and Intel/neural-chat-7b-v3-2. This merging strategy aims to combine the distinct capabilities of its constituent models, with specific weights applied during the merge process.

Key Capabilities & Performance

This model demonstrates solid performance across a range of benchmarks, as evaluated on the Open LLM Leaderboard. Its average score is 68.71, with notable results including:

• HellaSwag (10-Shot): 84.11
• Winogrande (5-shot): 78.53
• AI2 Reasoning Challenge (25-Shot): 66.38
• MMLU (5-Shot): 62.84
• TruthfulQA (0-shot): 63.59
• GSM8k (5-shot): 56.79

Prompting and Quantization

The model supports various prompt templates, with ChatML (from OpenHermes-2.5-Mistral-7B) and the neural-chat-7b-v3-2 format being recommended. For optimized deployment, several quantized versions are available through TheBloke, including GPTQ, GGUF, and AWQ formats, making it suitable for resource-constrained environments.