Qiskit/Qwen2.5-Coder-14B-Qiskit

Qwen-2.5-Coder-14B-Qiskit Overview

Qwen-2.5-Coder-14B-Qiskit is a specialized 14.7 billion parameter causal language model, built upon the Qwen2.5-Coder architecture. Developed by Qiskit, this model is specifically fine-tuned for Qiskit coding, ensuring compatibility with Qiskit version 2.0 APIs and syntax. It features a transformer architecture with RoPE, SwiGLU, RMSNorm, and Attention QKV bias, and supports an extensive context length of up to 131,072 tokens.

Key Capabilities & Features

• Enhanced Qiskit Code Performance: Demonstrates significant improvements in Qiskit code generation, reasoning, and fixing compared to previous Qiskit-specialized models.
• Long-Context Support: Capable of handling up to 128K tokens, with a full context length of 131,072 tokens, utilizing techniques like YaRN for extrapolation.
• Robust Foundation: Provides a comprehensive base for real-world applications such as Code Agents, while retaining strong general and mathematical competencies.
• Benchmark Performance: Achieves competitive scores on various benchmarks, including 49.01 on QiskitHumanEval, 91.46 on HumanEval, and 77.60 on MBPP, often outperforming other Qiskit-focused models in its class.

Ideal Use Cases

• Qiskit Code Generation: Generating quantum circuits and Qiskit-specific code snippets.
• Code Reasoning & Debugging: Assisting in understanding and fixing issues within Qiskit codebases.
• Code Agent Development: Serving as a core component for intelligent code agents focused on quantum computing tasks.
• Long-Form Code Analysis: Processing and generating code within very large contexts, beneficial for complex projects.