### Instruction: Summarize the key details of the Starlight model in a few sentences.

### Response: Starlight is a 13B parameter transformer model trained on the AverageData and Above the Clouds datasets for conversational text generation. It has strong language modeling capabilities but lacks true language understanding and may generate incorrect or biased text, so outputs should be monitored and safeguards implemented. The model is intended for use in chatbots and content creation applications.

Instructions for Safe Use

Avoid exposing Starlight to offensive, unethical, dangerous or illegal prompts
Monitor outputs for signs of bias, toxicity or factual incorrectness
Do not rely on Starlight for high-stakes or safety critical applications

Limitations

May hallucinate or generate incorrect information
Large model size leads to high compute requirements

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	46.87
ARC (25-shot)	59.3
HellaSwag (10-shot)	82.15
MMLU (5-shot)	55.67
TruthfulQA (0-shot)	37.39
Winogrande (5-shot)	76.64
GSM8K (5-shot)	10.84
DROP (3-shot)	6.08

NewstaR/Starlight-13B

Starlight (13B)

The model follows the Alpaca template:

Example:

Instructions for Safe Use

Limitations

Open LLM Leaderboard Evaluation Results

NewstaR/
Starlight-13B