Filtered Dataset Long: High-Quality Reasoning Traces for Language Model Fine-Tuning

A subset of rewritten reasoning traces from talzoomanzoo/rewritten_dataset_base, filtered for quality. The dataset was created by talzoomanzoo and last updated on 2026-05-22. It includes problem statements, original teacher reasoning, and reference answers, with rows retained only when judged suitable for supervised fine-tuning by a large language model.

Use Cases

• Supervised fine-tuning of language models based on high-quality reasoning traces.
• Training models to generate step-by-step solutions based on problem statements and teacher reasoning.
• Benchmarking model reasoning performance against filtered, high-scoring examples.
• Analyzing the characteristics of reasoning traces deemed suitable for instruction tuning.

Strengths

• Rows were filtered using a score threshold (≥7) from a large language model judge (openai/gpt-oss-120b via OpenRouter).
• Includes structured fields for problem statements, reasoning traces, and answers.
• Dataset version and progress are tracked via a separate dataset (talzoomanzoo/filtered_dataste_long_progress).

Limitations

• Row count, file formats, and license information are unknown.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect bias inherent to the source datasets and the judging model's preferences.

Provenance

Source

Subset of talzoomanzoo/rewritten_dataset_base, filtered via OpenRouter.

Collection Method

Rows were kept when a language model judge scored the rewritten trace as suitable for SFT (score ≥ 7).

Freshness

Last updated 2026-05-22 08:20:18; freshness should be verified.