Dracarys2-72B-Instruct Overview

Dracarys2-72B-Instruct is a 72.7 billion parameter instruction-tuned model developed by Abacus.AI, building upon the Qwen2.5-72B-Instruct base model. It is part of the "Smaug" series, with a specific focus on enhancing coding performance across various base models.

Key Capabilities & Performance

This model demonstrates notable improvements in coding benchmarks compared to its foundational model:

LiveCodeBench (LCB) Scores: Dracarys2-72B-Instruct shows higher scores across key LCB metrics:
- Code Generation: 53.80 (vs. 53.03 for Qwen2.5-72B-Instruct)
- Code Execution (COT): 89.12 (vs. 88.72)
- Test Output Prediction: 59.61 (vs. 46.28)
Code Generation Breakdown: It particularly excels in the "Easy" and "Medium" categories of Code Generation within LCB.
Test Output Prediction Breakdown: Significant improvements are observed across all difficulty levels (Easy, Medium, Hard) for Test Output Prediction.

Use Cases

Dracarys2-72B-Instruct is particularly well-suited for applications requiring robust code generation and understanding, especially when compared to its base model. Its enhanced performance on LiveCodeBench suggests strong capabilities for tasks such as:

Generating Python code, as demonstrated in the provided example.
Assisting with data science coding tasks using libraries like Pandas and NumPy.
Scenarios where accurate code execution and test output prediction are critical.