scribelm-dataset / README.md
gnumanth's picture
Upload 388-post Hemanth HM multi-task writing, rewrite, social & critic dataset
832feab verified
|
Raw History Blame Contribute Delete
872 Bytes
---
license: mit
task_categories:
- text-generation
- text-classification
language:
- en
size_categories:
- 1K<n<10K
---
# `gnumanth/scribelm-dataset` — Hemanth HM Technical Writing, Style-Transfer & Voice Critic Dataset
Built from **388 real Markdown blog posts (`~230,000` tokens)** and social adaptations in `~/labs/scribe` (`h3manth.com/scribe`), spanning 1,341 multi-task examples (`train.jsonl`: 1,072, `val.jsonl`: 123, `test.jsonl`: 146) across 4 tasks:
- `blog`: Technical blog post generation from title, category, tags, and summary
- `rewrite`: Corporate / dense technical draft $\rightarrow$ Hemanth HM practitioner voice
- `social`: LinkedIn, Hacker News, Reddit (`TL;DR:`), and X launch threads
- `critic`: Calibrated `p_hemanth` authenticity labels + 5-axis style telemetry (`directness`, `hook_strength`, `zero_fluff`, `code_first`, `cadence`)