Goedel-Prover-V2: Supervised Fine-Tuning Data for Formal Theorem Proving

Goedel-Prover-V2 SFT Dataset contains 1,745,010 samples for supervised fine-tuning in formal theorem proving. The dataset was created by Goedel-LM and is associated with a 2025 arXiv preprint. It was last updated on the Hugging Face platform in March 2026.

Use Cases

• Supervised fine-tuning of language models for formal theorem proving based on scaffolded data synthesis.
• Training models for mathematical reasoning and proof generation based on the described synthesis and self-correction methodology.
• Benchmarking AI systems on formal mathematics tasks based on the dataset's scale and purpose.
• Research into data synthesis and self-correction techniques for AI in mathematics based on the associated paper.

Strengths

• Large scale with 1,745,010 samples for training.
• Directly associated with a peer-reviewed arXiv preprint (arXiv:2508.03613, 2025).
• Specifically designed for supervised fine-tuning in a focused domain (formal theorem proving).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific model architectures.
• Description metadata is limited; actual data quality and sample structure require manual inspection after download.

Provenance

Source

Goedel-LM

Collection Method

Likely generated via scaffolded data synthesis and self-correction as described in the associated paper.

Freshness

Last updated 2026-03-02 19:27:28; freshness should be verified.