KletterMix-12B: German-Language Text Dataset with Cluster Labels

A German-language text dataset released by AIML-TUDA on June 4, 2026. It is a smaller subset of the KletterMix collection, intended for inspection and controlled experiments. Each row contains text alongside a cluster assignment, a GPT-2 token count, and a proxy score.

Use Cases

• Analyze text clustering patterns based on the provided cluster assignments.
• Benchmark tokenization efficiency using the included GPT-2 token counts.
• Filter or sample text data for experiments based on the proxy score.
• Inspect the characteristics of a curated subset of a larger German text collection.

Strengths

• Data includes structured metadata like cluster assignments and token counts per row.
• Dataset is explicitly designed as a compact release for easier inspection and analysis.
• Released as sharded JSONL files, which suggests a structured and potentially scalable format.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, total size, and license information are unknown, limiting suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

AIML-TUDA

Freshness

Last updated 2026-06-04 13:59:28; freshness should be verified.