TheBloke/CAMEL-13B-Combined-Data-fp16
TheBloke/CAMEL-13B-Combined-Data-fp16 is a 13 billion parameter chat large language model, fine-tuned by Camel AI from the LLaMA-13B architecture. It was trained on 229K CAMEL framework conversations, 100K ShareGPT conversations, and 52K Alpaca instructions. This model excels in conversational AI tasks, achieving an average score of 58.1 on the Open LLM Benchmark, outperforming LLaMA-30B and matching LLaMA-65B.
Loading preview...
Model Overview
CAMEL-13B-Combined-Data is a 13 billion parameter chat-optimized large language model developed by Camel AI. It is built upon the LLaMA-13B architecture and has been extensively fine-tuned using a diverse dataset to enhance its conversational capabilities.
Key Capabilities & Training
This model's training regimen involved a combination of 229,000 conversations generated through the proprietary CAMEL framework, 100,000 public English conversations from ShareGPT, and 52,000 instructions from the Alpaca dataset. This comprehensive training approach aims to provide robust performance across various conversational and instruction-following tasks.
Performance Highlights
Evaluated using EleutherAI's language model evaluation harness, CAMEL-13B-Combined-Data achieved an impressive average score of 58.1 on the Hugging Face Open LLM Benchmark. This places its performance on par with LLaMA-65B and notably surpasses LLaMA-30B, demonstrating strong capabilities for a 13B parameter model. Specific benchmark scores include:
- ARC-C: 55.5
- HellaSwag: 79.3
- MMLU: 50.3
- TruthfulQA: 47.3
Use Cases
This model is particularly well-suited for applications requiring advanced conversational AI, instruction following, and general language understanding, making it a strong candidate for chatbots, virtual assistants, and other dialogue-based systems.