MachineLearningLM: Synthetic Tabular ML Tasks from Structural Causal Models

MachineLearningLM provides a pretraining corpus for large language models designed to equip them with in-context machine learning capabilities. The dataset consists of millions of ML tasks synthesized from structural causal models, spanning various shot counts up to 1,024. It was created by MachineLearningLM and last updated on December 12, 2025.

Use Cases

• Pretraining LLMs for in-context learning based on the described synthetic ML tasks.
• Benchmarking LLM performance on structured causal inference problems based on the SCM-derived tasks.
• Studying few-shot learning capabilities on tabular data using the provided shot counts up to 1,024.

Strengths

• The corpus is synthesized from millions of structural causal models, suggesting a large scale of generation.
• It explicitly includes tasks spanning various shot counts up to 1,024, providing a range of learning examples.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The synthetic nature of the data may not fully reflect real-world data distributions.

Provenance

Source

MachineLearningLM

Collection Method

Synthesized from structural causal models.

Freshness

Last updated 2025-12-12 11:35:00