Dolphin 2.9 LLaMA3 70B
| Property | Value |
|---|---|
| Parameter Count | 70.6B |
| Model Type | Language Model |
| Architecture | LLaMA3-based |
| License | META LLAMA 3 COMMUNITY LICENSE |
| Training Hardware | 8xH100 GPU Node |
| Context Length | 8K tokens |
What is dolphin-2.9-llama3-70b?
Dolphin 2.9 is a sophisticated language model built on Meta's LLaMA3 70B architecture, fine-tuned through qLoRA for enhanced conversational AI capabilities. Developed by Eric Hartford, Lucas Atkins, and Fernando Fernandes, this model represents a significant advancement in uncensored AI technology, trained on a diverse dataset including CodeFeedback, OpenHermes, and UltraChat.
Implementation Details
The model was trained for 2.5 days using 8 H100 GPUs provided by Crusoe Cloud. It implements the ChatML prompt template format and maintains an 8K token context window. The model uses BF16 tensor type for optimal performance and memory efficiency.
- Comprehensive fine-tuning on 10 diverse datasets
- Implements ChatML prompt format for structured interactions
- Supports function calling capabilities
- Uses qLoRA fine-tuning methodology
Core Capabilities
- Advanced conversational abilities
- Code generation and analysis
- Uncensored response generation
- Mathematical problem-solving
- Agent-like behavior support
- Function calling integration
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its uncensored nature and comprehensive training across diverse datasets, making it highly adaptable while maintaining compliance with user requests. It's particularly notable for combining conversational abilities with coding expertise and function-calling capabilities.
Q: What are the recommended use cases?
The model is suited for various applications including conversational AI, code generation, mathematical problem-solving, and function-calling implementations. However, users should implement their own alignment layer before deploying it as a service due to its uncensored nature.
