Hugging Face Overview
Model Overview
M4-ai/NeuralReyna-Mini-1.8B-v0.3 is a 1.8 billion parameter language model that builds upon the aloobun/Reyna-Mini-1.8B-v0.2 base. It has been further fine-tuned using DPO (Direct Preference Optimization) with the argilla/OpenHermes2.5-dpo-binarized-alpha dataset, enhancing its performance across various domains. The model is trained to follow OpenAI's ChatML prompt format, making it compatible with common instruction-following paradigms.
Key Capabilities
- Multifaceted Skillset: Demonstrates capabilities in a range of areas including coding, mathematical problem-solving, scientific reasoning, and engaging in roleplay scenarios.
- Function Calling: Equipped with the ability to handle function calling, which is crucial for integrating with external tools and APIs.
- DPO Fine-tuning: Leverages Direct Preference Optimization for improved alignment and response quality.
Performance Benchmarks
Evaluated on the Open LLM Leaderboard, NeuralReyna-Mini-1.8B-v0.3 shows a balanced performance for its size:
- Avg. Score: 41.77
- MMLU (5-Shot): 44.22
- HellaSwag (10-Shot): 61.13
- AI2 Reasoning Challenge (25-Shot): 35.58
- GSM8k (5-Shot): 6.75
Good For
- Developers seeking a compact yet versatile model for applications requiring a mix of general language understanding and specialized tasks like code generation or function calling.
- Use cases where adherence to the ChatML prompt format is beneficial for integration.
- Experimentation with DPO-tuned models in a smaller parameter footprint.