Jongsim/GLM-5.1-Reasoning-1M-filtered
收藏Hugging Face2026-04-21 更新2026-04-26 收录
下载链接：
https://hf-mirror.com/datasets/Jongsim/GLM-5.1-Reasoning-1M-filtered
下载链接
链接失效反馈官方服务：
资源简介：
---
license: apache-2.0
task_categories:
- text-generation
language:
- en
- ko
- multilingual
tags:
- reasoning
- math
- science
- stem
- filtered
size_categories:
- 100K<n<1M
---
# GLM-5.1-Reasoning-1M-filtered
Refusal-filtered subset of `zai-org/GLM-5.1-Reasoning-1M`.
## Filtering
Conservative regex/string match for explicit refusal patterns (e.g., "I cannot",
"I can't", "I'm not able to", "as an AI", "죄송합니다", etc.) applied on the
assistant response field. Drop ratio: ~0.00% (very few refusals in source).
## Files
| File | Size | Records (kept) |
|---|---|---|
| `main.jsonl` | 19 GB | (full main subset, refusals removed) |
| `Math.jsonl` | 4.1 GB | mathematics reasoning |
| `PHD-Science.jsonl` | 3.8 GB | 103,705 |
| `Multilingual-STEM.jsonl` | 3.6 GB | 92,781 |
Total: 30 GB JSONL.
## Source
- Original: [zai-org/GLM-5.1-Reasoning-1M](https://huggingface.co/datasets/zai-org/GLM-5.1-Reasoning-1M)
- Filter script: `filter_glm_refusals.py` (sequential to avoid HDD I/O contention)
## License
Inherits Apache-2.0 from the source dataset.
提供机构：
Jongsim
