Model Overview
Locutusque/Hercules-4.0-Mistral-v0.2-7B is a 7 billion parameter language model, fine-tuned from Mistralai/Mistral-7B-v0.2. It is specifically engineered to improve instruction following, function calling, and conversational abilities within scientific and technical fields. The model was trained on the Hercules-v4.0 dataset, an expansion of OpenHermes-2.5, utilizing 700,000 examples over 8 Kaggle TPUs.
Key Capabilities
- Complex Instruction Following: Accurately executes multi-step instructions, including those with specialized terminology.
- Function Calling: Interprets and executes function calls, managing appropriate input and output values.
- Domain-Specific Knowledge: Engages in informative conversations across Biology, Chemistry, Physics, Mathematics, Medicine, and Computer Science.
- Training: Fine-tuned with a low learning rate (5e-06) using the Adam optimizer and a linear scheduler, without mixed precision, on bfloat16.
Intended Use Cases
- Specialized Chatbots: Ideal for developing knowledgeable conversational agents in scientific and technical domains.
- Instructional Assistants: Supports users with educational and step-by-step guidance across various disciplines.
- Code Generation and Execution: Facilitates code execution via function calls, aiding in software development and prototyping.
Performance Highlights
Evaluations show competitive performance, with an average score of 61.53 on the Open LLM Leaderboard. Specific metrics include:
- MMLU (5-shot): 62.66
- HellaSwag (10-shot): 82.60
- Winogrande (5-shot): 78.53
Limitations
Users should be aware of potential hallucinations and factual errors, especially in highly specialized areas. The model may also reflect biases present in its training data and has the potential for misuse due to its technical capabilities.