Claude Distills: Unified Datasets for Language Model Distillation

A curated collection of open-source datasets for distilling knowledge from Anthropic's Claude models. The repository contains at least two unified subsets, including 'claude-sonnet-4.6-120000x' with 119,446 samples and 'claude-opus-4.6-10000x' with 9,633 samples. The data was aggregated and formatted by ansulev, with credit to original creators, and was last updated on 2026-06-16.

Use Cases

• Train student models via knowledge distillation based on the described general, code, math, and psychology data.
• Benchmark distillation techniques using the unified and deduplicated datasets mentioned in the description.
• Fine-tune models for specific tasks like code generation or mathematical reasoning based on the described data subsets.

Strengths

• The repository provides at least 129,079 total samples across its listed subsets.
• The data has undergone a described process of unification and deduplication.
• The source of each subset is documented, with credit attributed to original creators.

Limitations

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

Provenance

Source

Aggregated from multiple open-source Claude distillation datasets.

Collection Method

Curated, unified, and deduplicated from original sources.

Freshness

Last updated 2026-06-16 16:25:36; freshness should be verified.