Model Overview

lgaalves/llama-2-13b-hf-platypus is a 13 billion parameter instruction fine-tuned language model built upon the LLaMA2 transformer architecture. Developed by Luiz G A Alves, this model is specifically trained on the garage-bAInd/Open-Platypus dataset, which focuses on STEM and logic-based content. The fine-tuning process utilized LoRA and was completed in approximately 2.5 hours on an A100-40GB GPU.

Key Capabilities & Performance

This model is designed for general instruction-following tasks, particularly excelling in areas requiring logical reasoning and factual recall due to its training data. While it shows strong performance, it is slightly behind the garage-bAInd/Platypus2-13B model on several benchmarks, including ARC, HellaSwag, MMLU, and TruthfulQA. However, it generally outperforms the base llama-2-13b-hf model across these metrics.

Training Details

• Base Model: LLaMA2-13B
• Fine-tuning Method: LoRA
• Training Dataset: garage-bAInd/Open-Platypus (STEM and logic-based)
• Language: English

Limitations

As with all large language models, this Llama 2 variant carries inherent risks. Its outputs cannot be predicted in advance and may occasionally produce inaccurate, biased, or objectionable responses. Developers are advised to conduct thorough safety testing tailored to their specific applications before deployment, as testing has primarily been in English and cannot cover all potential scenarios.