Dclm Baseline 1B: 1 Billion Token Web Content Sample

Dclm Baseline 1B is a 1 billion token sample created by codelion from the mlfoundations/dclm-baseline-1.0 dataset. It was generated using reservoir sampling to ensure statistical representativeness of the source's filtered, diverse web content. The dataset was last updated on November 2, 2025.

Use Cases

• Training language models on a representative sample of web text based on the described 1B token size.
• Conducting statistical analysis of web content characteristics based on the reservoir sampling methodology.
• Benchmarking model performance on filtered web data as described in the source dataset lineage.
• Ablation studies on training data scale and diversity based on the sampling approach.

Strengths

• Sample size of 1 billion tokens is explicitly stated.
• Created using reservoir sampling, a statistically unbiased random sampling algorithm.
• Derived from a known source dataset (mlfoundations/dclm-baseline-1.0).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

mlfoundations/dclm-baseline-1.0

Collection Method

Reservoir sampling from a source dataset of filtered, diverse web content.

Freshness

Last updated 2025-11-02 15:56:41; freshness should be verified.