Dataset · Text generation
threat-intelligence
by Muhammad Usman Siddiqui 316usman/threat-intelligence
A preference dataset for THREATINTELLIGENCE, harvested from real, human-labelled sources and curated by an automated harvesting harness with an LLM quality gate.
Dataset Card
A preference dataset for THREATINTELLIGENCE, harvested from real, human-labelled sources and curated by an automated harvesting harness with an LLM quality gate. Standard preference / DPO schema — each row: 80/10/10 train / validation / test (seeded shuffle): train:1632 / validation:204 / test:204 - synthetic:deepseek/deepseek-v4-flash-0731 prompt and chosen are real — harvested from the human-labelled sources listed above (accepted / upvoted / high-rated responses). rejected is partly synthetic. Most sources retain only the accepted answer, so where no genuine low-scored counterpart existed the negative was generated by defect injection: an LLM rewrites that row's own chosen with exactly…
Excerpt from the card by Muhammad Usman Siddiqui, licensed unknown.
Structure
default 2,040 rows
| Split | Rows | Size |
|---|---|---|
| train | 1,632 | 1.9 MB |
| validation | 204 | 293.7 KB |
| test | 204 | 295.4 KB |
Details
- Repository
- 316usman/threat-intelligence
- Publisher
- Muhammad Usman Siddiqui
- Task category
- Text generation
- Tags
- preference, dpo, reward-model
- Size category
- 1K<n<10K
- Languages
- dpo
- Revision
- e5613d57131c68a7231c6a0494dac3bdfe58ad24
- Last updated
- 2026-10-14
Files
5 files, 1.4 MB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| data/test-00000-of-00001.parquet | Data | 163.9 KB | 9d021b782989 |
| data/train-00000-of-00001.parquet | Data | 1.0 MB | 079e1b5222f4 |
| data/validation-00000-of-00001.parquet | Data | 164.6 KB | 0d5eaf2f8f21 |
| README.md | Documentation | 1.7 KB | — |
| .gitattributes | Repository | 2.5 KB | — |
License and Download
- License
- unknown
- Access
- No access gate
Released by Muhammad Usman Siddiqui through its official repository on Hugging Face.