This repository provides scripts and data format definitions for fine-tuning Qwen3-8B on the HydroBench hydrogen-energy Q&A benchmark.
HydroBench is a bilingual (Chinese/English) domain dataset focusing on hydrogen production, storage, transport, fuel cells, safety, systems, and AI. It is designed for:
- Teaching / exam question answering
- Domain-specific assistant training
- Evaluation of LLMs on hydrogen energy literacy
HydroBench is organized as bilingual Q&A pairs, each with metadata describing its module and subtopic.
Each example has (at least) the following fields:
id: unique identifier, e.g.A-ELC-04module: high-level module (A–E)subtopic: submodule shorthand, e.g.ELCquestion_zh,question_en: question in Chinese / Englishanswer_zh,answer_en: reference answer in Chinese / English- (optional) other fields such as
difficulty,source, etc., depending on your version
The dataset currently contains ≈648 Q&A pairs (Chinese + English), distributed as follows:
| Module | Description | #Q&A |
|---|---|---|
| A | Hydrogen classification and production | 135 |
| B | Storage and transportation | 130 |
| C | Fuel cells and applications | 150 |
| D | Safety, economics, and strategy | 148 |
| E | System integration, AI, and life-cycle analysis | 85 |
| Submodule | Description | #Q&A |
|---|---|---|
| A-CLS | Color / classification definitions | 7 |
| A-SMR | Fossil / blue hydrogen (gasification / reforming) | 32 |
| A-ELC | Water electrolysis (AWE / PEM / SOEC / AEM) | 65 |
| A-NEW | Emerging H₂ production (photo-, electro-, thermo-catalysis, etc.) | 31 |
| B-GAS | Gaseous storage & transport (high-pressure tanks / pipelines / blending) | 44 |
| B-LIQ | Liquid hydrogen | 42 |
| B-SOLC | Solid & chemical carriers (ammonia / LOHC / metal hydrides) | 44 |
| C-TRA | Transportation applications | 65 |
| C-IND | Industrial applications | 45 |
| C-PWR | Power generation & buildings (CHP, reversible fuel cells) | 40 |
| D-SAF | Safety and standards | 50 |
| D-ECO | Economics and cost | 40 |
| D-STR | Global strategy | 58 |
| E-AI | AI and digital twins | 15 |
| E-INT | Electricity–hydrogen–carbon system integration | 20 |
| E-LCA | Life cycle assessment (LCA) | 15 |
| E-BUS | Business models / policy (label name may be adjusted) | 15 |
| E-SUP | Supply chain / supporting systems (label name may be adjusted) | 20 |
You can select Chinese only, English only, or bilingual samples when constructing training data.