Xenith-3B

Xenith-3B is a fine-tuned language model based on the microsoft/Phi-3-mini-4k-instruct model. It has been specifically trained on the AlignmentLab-AI/alpaca-cot-collection dataset, which focuses on chain-of-thought reasoning and instruction following.

Model Overview

• Model Name: Xenith-3B
• Base Model: microsoft/Phi-3-mini-4k-instruct
• Fine-Tuned On: AlignmentLab-AI/alpaca-cot-collection
• Model Size: 3 Billion parameters
• Architecture: Transformer-based LLM

Training Details

• Objective: Fine-tune the base model to enhance its performance on tasks requiring complex reasoning and multi-step problem-solving.
• Training Duration: 10 epochs
• Batch Size: 8
• Learning Rate: 3e-5
• Optimizer: AdamW
• Hardware Used: 2x NVIDIA L4 GPUs

Performance

Xenith-3B excels in tasks that require:

• Chain-of-thought reasoning
• Instruction following
• Contextual understanding
• Complex problem-solving
• The model has shown significant improvements in these areas compared to the base model.