Qwen3-14B-algebraic-geometry-reasoning-lora

Model Description

This model is a fine-tuned version of unsloth/Qwen3-14B-unsloth-bnb-4bit (using unsloth/OpenMathReasoning-mini as a base alignment) specialized in Algebraic Geometry. It is designed to assist in complex mathematical reasoning, proof construction, and conceptual explanation within the field of modern algebraic geometry.

• Finetuning Data: High-quality synthetic data and structured reasoning paths derived from Andreas Gathmann's Algebraic Geometry Class Notes.
• Specialization: Covers Topics from classical algebraic varieties to the language of schemes, sheaves, and sheaf cohomology.
• Base Model: unsloth/Qwen3-14B
• Framework: Unsloth (Fast LoRA training)

Technical Background

This model attempts to bridge the gap between Large Language Models and formal algebraic geometry reasoning. By leveraging the structured proofs found in Gathmann's curriculum, the model is tuned to follow more rigorous logical steps when solving geometric problems.

Training Details

• Rank (R):32
• Alpha:32
• Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Data Source: Andreas Gathmann - Algebraic Geometry (Class Notes)

Intended Use

• Assistance in proof-writing for graduate-level algebraic geometry.
• Clarification of definitions (e.g., Separatedness of schemes, Proper morphisms).
• Reasoning exercises based on the OpenMathReasoning logic.

Try it in Colab

You may try this model in Google Colab using the following link: Colab.research.google.com. You may also find the corresponding .ipynb notebook file in this repository for direct use or adaptation.