LLM Reference

Platypus2 7B

About

The Platypus2-7B is a large language model based on the Llama 2-7B transformer architecture, specifically instruction-fine-tuned to excel in English language tasks. Developed by Cole Hunter and Ariel Lee, it was trained using a STEM and logic-focused dataset, allowing it to adeptly handle tasks that demand logical reasoning and problem-solving. Training was conducted on a single A100 80GB GPU utilizing LoRA for fine-tuning efficiency. Notable considerations include using fp16=False and bf16=True during fine-tuning for optimal performance. The model is licensed under a Non-Commercial Creative Commons license (CC BY-NC-4.0), and while it shows strengths in text generation and question answering, its abilities in other languages are uncertain. It shares common LLM risks of potentially generating biased or inaccurate outputs.

Capabilities

MultimodalFunction CallingTool UseJSON Mode

Specifications

FamilyPlatypus2
Parameters7B
ArchitectureDecoder Only
Specializationgeneral