MaziyarPanahi/phi-2-logical-sft
MaziyarPanahi/phi-2-logical-sft is a 3 billion parameter causal language model fine-tuned by MaziyarPanahi from Microsoft's phi-2 architecture. This model is specifically fine-tuned on the Open-Platypus dataset, enhancing its logical reasoning and instruction-following capabilities. It achieves an average score of 61.50 on the Open LLM Leaderboard, making it suitable for tasks requiring structured responses and logical problem-solving within its 2048 token context length.
Loading preview...
Overview
This model, MaziyarPanahi/phi-2-logical-sft, is a 3 billion parameter language model derived from Microsoft's phi-2. It has been fine-tuned by MaziyarPanahi using the Open-Platypus dataset, which is designed to improve instruction-following and logical reasoning. The model demonstrates a validation loss of 1.0075 after training for 2 epochs.
Key Capabilities
- Enhanced Logical Reasoning: Fine-tuning on
Open-Platypusdataset aims to improve the model's ability to follow instructions and perform logical tasks. - Instruction Following: Designed to generate structured and coherent responses based on given instructions, as shown in the provided examples.
- Accessibility: Quantized GGUF versions are available, promoting deployment on commodity hardware without specialized accelerators.
Performance Highlights
Evaluated on the Open LLM Leaderboard, the model achieved an average score of 61.50. Notable scores include 75.14 on HellaSwag (10-Shot) and 74.90 on Winogrande (5-shot), indicating strong performance in common sense reasoning. It also scored 55.80 on GSM8k (5-shot) for mathematical reasoning.
Good for
- Applications requiring logical problem-solving and step-by-step reasoning.
- Instruction-based text generation and question answering.
- Developers seeking a compact yet capable model for deployment on resource-constrained environments, thanks to available GGUF quantizations.