Dolly-v2-3b
| Property | Value |
|---|---|
| Base Model | Pythia-2.8b |
| Parameters | 2.8 billion |
| License | MIT |
| Training Data | databricks-dolly-15k |
| Language | English |
What is dolly-v2-3b?
Dolly-v2-3b is an instruction-following large language model developed by Databricks, fine-tuned on their proprietary dataset of 15,000 instruction/response pairs. This model represents a significant step in open-source AI, being one of the first commercially-viable instruction-tuned LLMs released under a permissive license.
Implementation Details
Built on EleutherAI's Pythia-2.8b architecture, this model has been specifically trained to follow various types of instructions across multiple domains. The model utilizes the transformers library and can be efficiently run with bfloat16 precision to reduce memory usage without significant quality impact.
- Supports both standalone instruction following and context-based responses
- Compatible with popular frameworks like LangChain
- Optimized for GPU deployment with device mapping capabilities
- Implements custom InstructionTextGenerationPipeline for enhanced performance
Core Capabilities
- Brainstorming and ideation
- Classification tasks
- Closed and open Question-Answering
- Content generation
- Information extraction
- Summarization
Frequently Asked Questions
Q: What makes this model unique?
Dolly-v2-3b stands out for being one of the first commercially-licensed instruction-tuned LLMs, making it suitable for business applications while maintaining open-source principles. Despite its relatively small size, it demonstrates surprising instruction-following capabilities.
Q: What are the recommended use cases?
The model is best suited for basic instruction-following tasks, particularly in commercial applications where licensing is a concern. However, users should be aware of its limitations with complex syntax, mathematical operations, and factual accuracy.
