README.md

---
language:
- id
license: apache-2.0
task_categories:
- text-generation
- question-answering
size_categories:
- 100M<n<1B
tags:
- math
- reasoning
- cot
- synthetic
- lumina
- foundations
- agentic
- enterprise-oop
- urandom-entropy
---

# Lumina-Math-Foundations-1B

**Lumina-Math-Foundations-1B** is an industrial-scale foundational mathematical reasoning dataset in Indonesian, built upon the V18.0 Adaptive OOP & Cryptographic Entropy Engine.

### Key Schema Characteristics:
- **Clean Text Fields (`problem`, `step_solving`, `answer`):** 100% pure human natural language and LaTeX math without hardcoded prompt tokens.
- **Adaptive Code Architecture (`code`, `tool_calls_json`):** High-difficulty problems feature mandatory Enterprise `@dataclass` / OOP Classes, while elementary problems dynamically balance between functional snippets and classes based on `os.urandom` hardware entropy.
- **Decoupled Tool Metadata (`tool_calls_json`, `tool_responses_json`, `verification_payload`):** Isolated JSON strings for agentic fine-tuning and programmatic RLVR validation.