AndreiSobo/pact-qwen-tutor
AndreiSobo/pact-qwen-tutor is a 7.6 billion parameter model, fine-tuned from Qwen 2.5 7B Instruct, specifically designed to function as a Socratic coding tutor for Python programming. It provides guiding hints and asks Socratic questions to help students debug their code without revealing direct solutions. This model excels at identifying common coding errors and fostering self-discovery in algorithmic problem-solving, making it ideal for educational applications.
Loading preview...
PACT: Personalised AI Coding Tutor
PACT (Personalised AI Coding Tutor) is a specialized language model developed by Andrei Sobo, fine-tuned from the Qwen 2.5 7B Instruct base model. Its core purpose is to act as a Socratic tutor for Python programming, guiding students through debugging and problem-solving with hints and questions rather than direct answers.
Key Capabilities & Features
- Socratic Tutoring: Trained to ask guiding questions and provide hints, encouraging students to discover solutions independently.
- Error Identification: Capable of identifying various types of coding errors, including logic errors, edge case failures, and off-by-one errors.
- Supportive Tone: Designed to be encouraging and supportive, fostering a positive learning environment.
- Python-focused: Optimized for Python programming problems, particularly those found in algorithmic challenges like LeetCode.
- Controlled Output: Aims for a low "Code Leakage Rate" (<5%) and a high "Guiding Question Rate" (>70%) to ensure effective tutoring.
Training Details
The model was fine-tuned using QLoRA on a dataset of 227 synthetic examples of coding errors, generated using Claude Sonnet 4.5 and validated with GPT-5.2. This dataset focused on common student mistakes in algorithmic problems.
Intended Use Cases
PACT is primarily intended for educational purposes to:
- Help students learn Python through guided discovery.
- Assist in debugging common coding errors.
- Encourage critical thinking and problem-solving skills.
It is not designed for production code generation or automated grading systems. Users should ensure their input queries follow a specific structure, including problem description, student code, issue description, and a request for a hint, to achieve optimal results.