Qiskit/granite-3.3-8b-qiskit
The granite-3.3-8b-qiskit is an 8 billion parameter model developed by IBM Quantum & IBM Research, extended and fine-tuned from granite3.3-8b-base. It is specifically optimized for generating high-quality, non-deprecated Qiskit code, ensuring compatibility with Qiskit version 2.0 APIs and syntax. This model excels at assisting both experienced and new Qiskit users in building quantum computing code and responding to related instructions. Its training data includes publicly available code and synthetic data, with rigorous deduplication and filtering for PII and malware.
Loading preview...
Overview
The granite-3.3-8b-qiskit is an 8 billion parameter model developed by IBM Quantum & IBM Research. It is an extension and fine-tuning of the granite3.3-8b-base model, specifically designed to enhance its capabilities in generating high-quality and non-deprecated Qiskit code. The model was trained using Qiskit code and instruction data, ensuring compatibility with Qiskit version 2.0 APIs and syntax. Training data includes publicly available code (e.g., GitHub) and synthetic data, with strict filtering for licenses (Apache 2.0, MIT, Unlicense, Mulan PSL v2, BSD-2, BSD-3, CC BY 4.0), deduplication, and PII/malware filtering.
Key Capabilities
- Qiskit Code Generation: Specialized in producing functional and up-to-date Qiskit code for quantum computing tasks.
- Instruction Following: Capable of responding to Qiskit coding-related instructions and questions.
- Compatibility: Trained with Qiskit version 2.0, ensuring modern API and syntax adherence.
Benchmarks and Performance
While not leading in all general coding benchmarks, granite-3.3-8b-qiskit shows competitive performance in quantum-specific evaluations:
- QiskitHumanEval-Hard: 14.57
- QiskitHumanEval: 27.15
- HumanEval: 62.80
Intended Use
This model is ideal for:
- Quantum Computing Practitioners: As an assistant for building complex Qiskit circuits and programs.
- New Qiskit Users: To help in learning and generating correct Qiskit code, reducing the learning curve.
Limitations and Ethical Considerations
As with all LLMs, generated code is not guaranteed to be perfect. The model has not undergone specific safety alignment and may produce problematic outputs. Users are urged to use the model responsibly and ethically, especially given the ongoing research into hallucination and memorization in smaller models.