KIMI-K2.5-550000x: 550,000 Reasoning Traces from a Language Model

550,000 reasoning traces were distilled from the KIMI-K2.5 language model on high-reasoning tasks. The collection includes 2 billion tokens and is distributed across coding (60%), science (15%), math (10%), computer science (5%), logical questions (5%), and creative writing (5%). It was created by ansulev and last updated on Hugging Face in April 2026.

Use Cases

• Training or fine-tuning language models for code generation based on the 60% coding subset.
• Benchmarking model reasoning capabilities on science problems using the Physics, Chemistry, and Biology traces.
• Studying step-by-step logical reasoning processes for math and logical questions.
• Analyzing the structure of model-generated reasoning traces across different domains like creative writing.

Strengths

• Large scale with 550,000 distinct reasoning traces.
• Broad domain coverage across six distinct categories, with coding being the largest at 60%.
• Substantial token volume of 2 billion tokens for training or analysis.

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

Distilled from the KIMI-K2.5 language model.

Collection Method

Collected using a modified Datagen tool, as referenced in the description.

Freshness

Last updated 2026-04-03 09:49:47; freshness should be verified.