Hugging Face Reyna-Mini-1.8B-v0.2: A Compact Chat Model
aloobun/Reyna-Mini-1.8B-v0.2 is a 1.8 billion parameter language model, building upon the Qwen1.5-1.8B-Chat architecture. It has been fine-tuned using Supervised Fine-Tuning (SFT) on the Hercules v3 dataset, marking it as the third model in this development series. The model is specifically formatted to work with ChatML, making it suitable for conversational AI applications.
Key Capabilities & Performance
This model demonstrates a balanced performance across several benchmarks, with an overall average score of 45.94. Notable scores include:
- MMLU (5-Shot): 44.75
- HellaSwag (10-Shot): 60.19
- TruthfulQA (0-shot): 41.24
- GSM8k (5-shot): 31.31
These results suggest its utility in general reasoning, common sense, and question-answering tasks, particularly within a chat context. The developers plan to further enhance the model through DPO (Direct Preference Optimization) training.
Ideal Use Cases
- Chatbot Development: Its ChatML format and fine-tuning make it well-suited for creating interactive conversational agents.
- General-Purpose Assistant: Capable of handling a variety of queries, from factual questions to creative prompts, as indicated by its benchmark performance.
- Resource-Constrained Environments: As a 1.8B parameter model, it offers a balance of capability and efficiency for deployment where larger models might be impractical.