Tumpeng 1: 48.6K Indonesian Question-Answer Pairs for Language Model Fine-Tuning

Indonesian synthetic dataset containing 48.6K input-output pairs, totaling 14.8 million words. It was created by afrizalha using Llama 3 70B and released as a research preview on Hugging Face in November 2024. The dataset is intended to fine-tune Llama 3 8B models for improved Indonesian language capabilities.

Use Cases

• Fine-tuning language models for Indonesian question answering based on the 48.6K input-output pairs.
• Benchmarking model performance on Indonesian language tasks based on the synthetic question-answer format.
• Training models to generate Indonesian text responses based on the described input-output structure.

Strengths

• Contains 48.6K structured input-output pairs, providing a substantial volume of training examples.
• Totals 14.8 million words, indicating significant linguistic content for model training.

Limitations

• Dataset is a research preview and is explicitly not curated for factual accuracy or safety.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Synthetic data generated by afrizalha using Llama 3 70B.

Collection Method

Synthetically generated.

Freshness

Last updated 2024-11-02 06:44:10