NewstaR/Starlight-7B
NewstaR/Starlight-7B is a 7 billion parameter transformer model developed by NewstaR, trained on the AverageData and Above the Clouds datasets. It is designed for conversational text generation, following the Alpaca instruction template. The model demonstrates strong language modeling capabilities, achieving an average score of 54.3 across various benchmarks, including 78.57 on HellaSwag and 46.8 on MMLU. Its primary application is in chatbots and content creation, though users should be aware of its limitations regarding factual accuracy and potential biases.
Loading preview...
Starlight-7B Overview
NewstaR/Starlight-7B is a 7 billion parameter transformer model developed by NewstaR, specifically trained for conversational text generation. It utilizes the Alpaca instruction template, making it suitable for various dialogue-based applications. The model was trained on the AverageData and Above the Clouds datasets, contributing to its language modeling capabilities.
Key Capabilities & Performance
Starlight-7B demonstrates solid performance across several language understanding benchmarks, with an overall average score of 54.3. Notable benchmark results include:
- ARC (25-shot): 53.07
- HellaSwag (10-shot): 78.57
- MMLU (5-shot): 46.8
- TruthfulQA (0-shot): 38.75
These scores indicate its proficiency in tasks requiring common sense reasoning, natural language inference, and general knowledge, though it may struggle with complex reasoning or factual accuracy.
Use Cases & Limitations
Starlight-7B is primarily intended for use in chatbots and content creation applications where conversational text generation is required. Users should be aware of its limitations, which include the potential for hallucination, generation of incorrect or biased information, and high compute requirements due to its size. It is not recommended for high-stakes or safety-critical applications, and outputs should always be monitored for accuracy and appropriateness.