Datasetpublic

cloverx-id/Lumina-Math-Foundations-1B

Lumina-Math-Foundations-1B Lumina-Math-Foundations-1B is an industrial-scale foundational mathematical reasoning dataset in Indonesian, Dataset Summary Language: Indonesian (id) with LaTeX mathematical formulas. Scale: 1 Billion Synthetic High-Fidelity Mathematical Reasoning instances.--- Data Schema & Field Breakdown Field Name Type Description problem string 100% pure human natural language problem statement with standard LaTeX math… See the full description on the dataset page: https://huggingface.co/datasets/cloverx-id/Lumina-Math-Foundations-1B.

sourceHugging Faceapache-2.0updated 2d agoView on Hugging Face

2likes1.4kdownloads

Dataset Card

Lumina-Math-Foundations-1B

Lumina-Math-Foundations-1B is an industrial-scale foundational mathematical reasoning dataset in Indonesian,

Dataset Summary

—Language: Indonesian (id) with LaTeX mathematical formulas.
—Scale: 1 Billion Synthetic High-Fidelity Mathematical Reasoning instances.---

Data Schema & Field Breakdown

Field Name	Type	Description
`problem`	`string`	100% pure human natural language problem statement with standard LaTeX math formatting.
`step_solving`	`string`	Step-by-step rigorous logical Chain-of-Thought (CoT) mathematical reasoning.
`answer`	`string`	The extracted canonical final answer or solution value.
`code`	`string`	Complete Python implementation. Elementary problems use dynamic functional code; high-tier problems strictly feature Enterprise `@dataclass` OOP architectures.
`tool_calls_json`	`string`	Isolated JSON payload specifying function/tool invocation schemas for agentic workflows.
`tool_responses_json`	`string`	Execution results from external interpreters/solvers.
`verification_payload`	`string`	Deterministic verification criteria for programmatic Reinforcement Learning with Verifiable Rewards (RLVR).

Recommended SFT Prompt Formatting

1. Chain-of-Thought (CoT) Mode (ChatML)

<|im_start|>user
{problem}<|im_end|>
<|im_start|>assistant
<thought>
{step_solving}
</thought>
Jawaban akhir: {answer}<|im_end|>

2. Agentic Code-Interpreter Mode

<|im_start|>user
{problem}<|im_end|>
<|im_start|>assistant
<thought>
{step_solving}
</thought>
{code}
<|im_end|>

License & Citation

This dataset is released under the Apache-2.0 License.

bibtex

@dataset{luminamath1b2026,
  author       = {{Silver Moon (cloverxion)} and {Earl Pan (earlpan-id)}},
  title        = {{Lumina-Math-Foundations-1B: Industrial Foundational Mathematical Reasoning Dataset}},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/cloverx-id/Lumina-Math-Foundations-1B}},
  url          = {https://huggingface.co/datasets/cloverx-id/Lumina-Math-Foundations-1B},
  note         = {Hugging Face Dataset}
}