GLM-5.2 Science-50000x: 50,000 Reasoning Traces for Physics, Chemistry, Biology

50,000 reasoning traces distilled from the GLM-5.2 model on high-difficulty science problems. The dataset contains 160 million tokens covering physics, chemistry, and biology topics. It was created by author 'ianncity' and last updated on July 12, 2026.

Use Cases

• Training language models for scientific reasoning based on the distilled traces.
• Benchmarking model performance on physics, chemistry, and biology problems.
• Analyzing the reasoning patterns of large models on high-difficulty science questions.
• Fine-tuning models for educational or tutoring applications in STEM fields.

Strengths

• Contains 50,000 individual reasoning traces.
• Covers three major scientific domains: physics, chemistry, and biology.
• Comprises 160 million tokens of distilled model output.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description notes potential prompt overlap with another dataset, which may affect uniqueness.

Provenance

Source

huggingface

Collection Method

Traces distilled from the GLM-5.2 model.

Freshness

Last updated 2026-07-12 17:44:42; freshness should be verified.