MiroMind-M1: 719K Problems for Mathematical Reasoning Model Training

MiroMind-M1 is a fully open-source series of reasoning language models built on Qwen-2.5. The dataset contains 719,000 curated problems used for supervised fine-tuning, with an additional 62,000 challenging examples used for reinforcement learning. It was created by miromind-ai and last updated on July 22, 2025.

Use Cases

• Fine-tuning language models for mathematical problem-solving based on the 719K supervised fine-tuning examples.
• Evaluating model performance on advanced math benchmarks like AIME24 and AIME25 as referenced in the description.
• Training models using reinforcement learning with verifiable rewards based on the 62K challenging examples.
• Studying context-aware multi-stage policy optimization methods for reasoning models.

Strengths

• Contains 719,000 curated problems for supervised fine-tuning.
• Includes 62,000 challenging examples for reinforcement learning with verifiable rewards.
• Model performance is documented on specific benchmarks (AIME24 and AIME25).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the specific dataset file is unknown, which may limit suitability assessment.

Provenance

Source

miromind-ai

Collection Method

Curated problems used for supervised fine-tuning and reinforcement learning of language models.

Time Range

Performance evaluated on AIME24 and AIME25 benchmarks.

Freshness

Last updated 2025-07-22 01:44:22; freshness should be verified.

Geography

null