TinyLlama-1.1bee is a 1.1 billion parameter causal language model developed by BEE-spoke-data, fine-tuned from PY007/TinyLlama-1.1B-intermediate-step-240k-503b. This model specializes in apiculture-related topics, providing information and insights on bees, hive management, and honey production. With a 2048-token context length, it serves as an educational resource for beekeeping enthusiasts and researchers.
Loading preview...
Model Overview
TinyLlama-1.1bee is a 1.1 billion parameter language model developed by BEE-spoke-data, fine-tuned specifically on a dataset related to bees and beekeeping (BEE-spoke-data/bees-internal). It is based on the PY007/TinyLlama-1.1B-intermediate-step-240k-503b architecture and aims to provide specialized knowledge in apiculture.
Key Capabilities
- Educational Engagement: Serves as an informative resource for novice beekeepers, enthusiasts, and those interested in understanding bees.
- General Queries: Answers questions about hive management, bee species, and honey extraction.
- Academic & Research Inspiration: Offers preliminary insights and ideas for studies in apiculture and environmental science.
Performance & Limitations
The model achieved an evaluation loss of 2.4285 and an accuracy of 0.4969 on its evaluation set. Its perplexity was measured at 11.3391. While specialized, it is not a certified apiculturist and should not be used for serious decisions regarding hive management or bee health. Users should always double-check information. The model's licensing follows Apache-2.0.
Open LLM Leaderboard Evaluation
On the Open LLM Leaderboard, TinyLlama-1.1bee achieved an average score of 29.15, with specific scores including ARC (25-shot) at 30.55, HellaSwag (10-shot) at 51.8, and MMLU (5-shot) at 24.25. GSM8K (5-shot) scored 0.23, indicating limited mathematical reasoning capabilities.