Hunter Alpha Reasoning Traces For Instruction Tuning

220,000 reasoning traces distilled from the Hunter Alpha model via OpenRouter. The dataset contains 800 million tokens and was created by user 'ianncity', with a last recorded update on March 15, 2026. It is composed of Math (30%), Coding (30%), Science (15%), Computer Science (15%), and Creative Writing (10%) content.

Use Cases

• Fine-tuning language models on high-quality reasoning traces from the 'Math' and 'Science' domains to improve logical problem-solving.
• Training models for code generation and understanding using the 'Coding' subset which includes traces for languages like C++, Java, and Rust.
• Augmenting creative writing models with the structured reasoning patterns from the 'Creative Writing' portion of the traces.
• Studying the distillation process and model reasoning patterns across the specified 'Computer Science' and general knowledge tasks.

Strengths

• Contains 220,000 distinct reasoning traces.
• Comprises 800 million tokens of text data.
• Covers a multi-domain distribution including Math, Coding, Science, and Creative Writing.

Limitations

• Exact row count and column structure are unknown, complicating direct analysis.
• Potential bias towards topics and problem types available on OpenRouter during the collection period.
• The dataset's collection method may introduce noise or stylistic artifacts from the source model.

Provenance

Source

Distilled from the Hunter Alpha model via the OpenRouter platform.

Collection Method

Reasoning traces were collected and distilled from model interactions.

Time Range

Collection period not specified; dataset snapshot from around March 2026.

Freshness

Last updated March 15, 2026; creator indicates ongoing updates until the source model is removed from OpenRouter.