AI-MO/Kimina-Prover-Preview-Distill-1.5B

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Apr 9, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

AI-MO/Kimina-Prover-Preview-Distill-1.5B is a 1.5 billion parameter theorem proving model developed by Project Numina and Kimi teams. It is a distillation of Kimina-Prover-Preview, trained via large-scale reinforcement learning, and is specifically optimized for competition-style problem solving in Lean 4. The model achieves state-of-the-art results on MiniF2F-test within its model size and compute budget, making it highly effective for formal reasoning tasks.

Loading preview...

Model Overview

AI-MO/Kimina-Prover-Preview-Distill-1.5B is a specialized 1.5 billion parameter model designed for theorem proving, developed collaboratively by Project Numina and Kimi teams. This model is a distilled version of the larger Kimina-Prover-Preview, which was trained using extensive reinforcement learning techniques.

Key Capabilities

  • Theorem Proving: Excels at formal reasoning and competition-style problem solving, particularly within the Lean 4 proof assistant environment.
  • State-of-the-Art Performance: Achieves leading results on the MiniF2F-test benchmark for its size and computational resources, demonstrating strong capabilities in mathematical theorem proving.
  • Distilled Architecture: Benefits from the advanced training of its larger predecessor, offering efficient performance for its parameter count.

Good For

  • Lean 4 Development: Ideal for developers and researchers working with Lean 4 who need assistance in formalizing and proving mathematical theorems.
  • Automated Reasoning: Suitable for applications requiring automated theorem proving, especially in competitive or research contexts.
  • Educational Tools: Can be integrated into tools for teaching and learning formal methods and proof assistants.