Dracarys2-72B-Instruct
| Property | Value |
|---|---|
| Parameter Count | 72.7B |
| License | Tongyi-Qianwen |
| Base Model | Qwen2.5-72B-Instruct |
| Developer | Abacus.AI |
| Tensor Type | BF16 |
What is Dracarys2-72B-Instruct?
Dracarys2-72B-Instruct is a specialized large language model that represents the latest advancement in the Smaug series, specifically designed to enhance coding performance. Built upon the Qwen2.5-72B-Instruct architecture, this model has been fine-tuned to achieve superior results in code generation and execution tasks.
Implementation Details
The model maintains compatibility with the Qwen2.5-72B-Instruct prompt format while delivering improved performance across various coding metrics. It utilizes the transformers library and supports text generation with customizable parameters such as temperature and top-p sampling.
- Implements BF16 precision for efficient computation
- Supports chat-based interactions through a structured message format
- Includes built-in safety measures and specialized tokens for generation control
Core Capabilities
- Superior code generation with 53.80% accuracy on LiveCodeBench
- Enhanced code execution with 89.12% success rate in Chain-of-Thought scenarios
- Improved test output prediction (59.61% accuracy)
- Exceptional performance on easy (88.79%) and medium (50.28%) difficulty coding tasks
Frequently Asked Questions
Q: What makes this model unique?
The model distinguishes itself through significantly improved LiveCodeBench scores compared to its base model, particularly in test output prediction where it shows up to 15% improvement in accuracy across all difficulty levels.
Q: What are the recommended use cases?
The model excels in code generation tasks, particularly for data science applications using Python, Pandas, and Numpy. It's especially effective for easy to medium difficulty coding challenges and can be integrated into development workflows for code generation and testing.
