Datasets:
metadata
language:
- he
- en
license: cc-by-4.0
tags:
- legal
- hebrew
- israel
- law
- court-rulings
- contracts
- fine-tuning
- instruction-tuning
task_categories:
- text-generation
- question-answering
size_categories:
- 10K<n<100K
BrainboxAI/legal-training-il
Hebrew & English Legal Training Dataset
17,613 instruction-tuning samples for training legal AI models on Israeli law.
Dataset Composition
| Source | Samples | Language | Content |
|---|---|---|---|
| Israeli Court Rulings (HF) | 3,978 | Hebrew | Supreme Court, Family, Criminal, Civil |
| Israeli Court Rulings (existing) | 3,982 | Hebrew | Various courts and tribunals |
| Kol-Zchut (כל-זכות) | 2,353 | Hebrew | Citizens' rights explanations |
| Open Law Book (ספר החוקים הפתוח) | 300 | Hebrew | Israeli legislation via Wikisource |
| Contract Clauses (CUAD-based) | 7,000 | English | 41 contract types, 28 clause categories |
| Total | 17,613 | 60% Hebrew |
Format
Alpaca instruction format (JSONL):
{
"instruction": "מה הזכויות שלי בנושא פיצויי פיטורים?",
"input": "",
"output": "לפי חוק פיצויי פיטורים, תשכ\"ה-1967..."
}
Sources
Hebrew Legal Data
- Israeli Court Rulings - guychuk/case-law-israel - 10,558 judgments from Israeli courts
- Kol-Zchut - kolzchut.org.il - Israeli citizens' rights knowledge base, 2,353 articles covering labor law, housing, health, insurance, disability, pensions, and more
- Open Law Book - openlaw.org.il - 5,969 consolidated Israeli laws via Hebrew Wikisource, a joint project of the Public Knowledge Workshop and Wikimedia Israel
English Legal Data
- Contract Clauses - Based on CUAD methodology, covering 41 contract types across 28 clause categories including License Grant, Liability Cap, Audit Rights, Anti-Assignment, Termination, Change of Control, and more
Instruction Types
| Type | Example | Source |
|---|---|---|
| Court ruling analysis | "נתח את פסק הדין הבא..." | Court rulings |
| Rights Q&A | "מה הזכויות שלי בנושא...?" | Kol-Zchut |
| Law explanation | "הסבר את החוק הבא בצורה מובנת..." | Open Law Book |
| Contract clause analysis | "What are the key legal implications of this clause?" | CUAD contracts |
Usage
from datasets import load_dataset
dataset = load_dataset("BrainboxAI/legal-training-il", split="train")
print(f"Samples: {len(dataset)}")
Fine-tuning with Unsloth
from unsloth import FastLanguageModel
from datasets import load_dataset
model, tokenizer = FastLanguageModel.from_pretrained(
model_name="unsloth/gemma-4-E2B-it",
max_seq_length=2048,
load_in_4bit=True,
)
dataset = load_dataset("BrainboxAI/legal-training-il", split="train")
Trained Model
This dataset was used to train BrainboxAI/law-il-E2B - a Hebrew legal AI model based on Gemma 4 E2B.
Limitations
- Court ruling outputs are currently summaries, not full legal analysis
- Contract clause data is in English with template-based analysis
- No case law citations or cross-references between sources
- Kol-Zchut content reflects rights information, not legal rulings
License
CC-BY-4.0. Individual sources may have additional terms:
- Kol-Zchut: CC-BY-SA
- Open Law Book: Public domain (Wikisource)
- Court rulings: Israeli government public data
Built by BrainboxAI