---
license: cc-by-4.0
task_categories:
  - text-generation
language:
  - en
  - zh
tags:
  - chemistry
  - cheminformatics
  - formulation
  - information-extraction
  - synthetic
size_categories:
  - n<1K
pretty_name: Formulation Records Instruct
---

# Formulation Records Instruct

A **synthetic, de-identified** instruction dataset for teaching an LLM to structure
and reason over semiconductor wet-chemistry formulation records. Three tasks:
`extract` (messy text → structured JSON), `normalize` (name/unit → canonical),
`explain` (optimizer result → plain-language explanation).

Built by [Formulith](https://www.formulith.com). Generator & schema:
<https://github.com/formulith/formulation-records-dataset>.

> 本專案部分研發由**數位發展部數位產業署 115 年 AI 算力平台**支持。
> Part of this work is supported by the **AI Computing Platform (2026), Administration for Digital Industries, Ministry of Digital Affairs, Taiwan**.


## Data is synthetic

Procedurally generated — **no real customer data, no proprietary formulations**.
Values are illustrative. Schema mirrors a real formulation-optimization pipeline
(molecule identity, normalized units, objectives, feasibility).

## Fields

See `schema.json` in the GitHub repo. Each row: `{ "task", "input", "output" }`.

## License

CC-BY-4.0.