Japanese-English Code Generation Dataset for Instruction Tuning

LLM-jp, a collaborative Japanese project, created this synthetic dataset for instruction tuning. It contains a subset of the 801,000-instruction Aratako/Synthetic-JP-EN-Coding-Dataset. The dataset was last updated in January 2025.

Use Cases

• Fine-tuning models to generate Python code from Japanese instructions using instruction-response pairs.
• Training models to translate English coding instructions into Japanese using bilingual text fields.
• Evaluating model performance on cross-language code synthesis tasks using synthetic prompts.
• Improving instruction-following capability for coding tasks via structured synthetic examples.

Strengths

• Derived from a source dataset containing 801,000 instructions.
• Created by a named collaborative research project (LLM-jp).
• Specifically designed for instruction tuning applications.

Limitations

• The exact size, row count, and column structure of this subset are unknown.
• Data is synthetic, generated by models, which may introduce patterns not found in human-written code.
• Limited provenance details; original collection method and time range are unspecified.

Provenance

Source

LLM-jp collaborative project.

Collection Method

Synthetic generation; subset of Aratako/Synthetic-JP-EN-Coding-Dataset-801k.

Freshness

Last updated on January 31, 2025.