Model Overview

kyujinpy/SOLAR-Platypus-10.7B-v1 is a 10.7 billion parameter auto-regressive language model developed by Kyujin Han. It is based on the Llama2 architecture and was fine-tuned from the upstage/SOLAR-10.7B-v1.0 base model. The training utilized LoRA (Low-Rank Adaptation) with specific configurations (LoRA_r: 16, LoRA_alpha: 16, LoRA_dropout: 0.05) on the kyujinpy/Open-platypus-Commercial dataset, using an Alpaca prompt template.

Key Capabilities & Performance

This model is designed for general text generation tasks. Its performance is evaluated on the Open LLM Leaderboard, where it achieved an average score of 58.62. Specific benchmark results include:

• ARC (25-Shot): 61.69
• HellaSwag (10-Shot): 84.23
• MMLU (5-Shot): 60.37
• TruthfulQA (0-shot): 51.58
• Winogrande (5-shot): 82.79
• GSM8k (5-shot): 11.07

When compared to its base model, upstage/SOLAR-10.7B-v1.0, this fine-tuned version shows variations in performance across different benchmarks, with the base model generally scoring higher on average (66.04) and significantly higher on GSM8K.

Good for

• Developers looking for a 10.7B parameter model fine-tuned on a commercial-friendly dataset.
• Experimentation with models based on the Llama2 architecture and LoRA fine-tuning.
• Applications requiring text generation where the specific benchmark scores align with task requirements.