hydra-project/OpenHercules-2.5-Mistral-7B
OpenHercules-2.5-Mistral-7B is a 7 billion parameter language model developed by hydra-project, created by merging Locutusque/Hercules-2.5-Mistral-7B and teknium/OpenHermes-2.5-Mistral-7B. This model leverages the Mistral architecture with a 4096-token context length, offering a balanced performance across various reasoning and language understanding tasks. It is particularly well-suited for general-purpose conversational AI and text generation where a blend of logical coherence and broad knowledge is beneficial.
Loading preview...
OpenHercules-2.5-Mistral-7B Overview
OpenHercules-2.5-Mistral-7B is a 7 billion parameter language model built upon the Mistral architecture, developed by hydra-project. This model is a strategic merge of two distinct base models: Locutusque/Hercules-2.5-Mistral-7B and teknium/OpenHermes-2.5-Mistral-7B, utilizing the slerp merge method via LazyMergekit. This merging approach aims to combine the strengths of both foundational models, resulting in a versatile and capable language model.
Key Capabilities & Performance
Evaluations on the Open LLM Leaderboard indicate a strong average performance, with specific scores highlighting its proficiency across various benchmarks:
- Avg. Score: 66.55
- AI2 Reasoning Challenge (25-Shot): 64.25
- HellaSwag (10-Shot): 84.84
- MMLU (5-Shot): 64.21
- Winogrande (5-shot): 78.93
- GSM8k (5-shot): 59.21
These metrics suggest a model that performs competently in reasoning, common sense, and general knowledge tasks, making it a robust choice for a variety of applications. Its 4096-token context window supports moderate-length interactions.
Use Cases
OpenHercules-2.5-Mistral-7B is well-suited for:
- General-purpose conversational agents: Its balanced performance makes it effective for chatbots and interactive AI.
- Text generation: Capable of producing coherent and contextually relevant text for various prompts.
- Reasoning tasks: Demonstrates solid performance in logical and common-sense reasoning benchmarks.
- Instruction following: The merged nature, including OpenHermes, suggests good instruction-following capabilities.