TESSY-Math-12K: 12,000 Synthetic Programming Contest Problems for Model Fine-Tuning

12,000 programming contest training problems synthesized for the Qwen3-8B model. The dataset was created by CoopReason using a teacher-student cooperation framework with GPT-OSS-120B as the teacher model to preserve strong reasoning capabilities. It was last updated on April 17, 2026.

Use Cases

• Fine-tuning language models for mathematical reasoning based on the described programming contest problems.
• Training models on synthetic data aligned with a teacher model's reasoning capabilities as described in the paper.
• Studying teacher-student knowledge transfer frameworks for generating student-consistent data.
• Benchmarking model performance on programming contest-style problems.

Strengths

• Dataset size is explicitly 12,000 entries as indicated in the title.
• Created using a documented teacher-student cooperation framework described in an associated paper.
• Specifically designed for fine-tuning the Qwen3-8B model.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count beyond the title's 12K is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

CoopReason

Collection Method

Synthesized using a teacher-student cooperation framework with GPT-OSS-120B as the teacher model.

Freshness

Last updated 2026-04-17 03:18:08; freshness should be verified.