Hinglish Instruct: 10,378 Romanized Hindi-English Dialogue Pairs

Akshar Hinglish Instruct is a code-mixed Romanized Hindi-English instruction-tuning dataset containing 10,378 dialogue pairs. It was created by Sujalvc and last updated on Hugging Face in May 2026. The dataset is designed to train conversational models on natural, domain-diverse South Asian speech patterns.

Use Cases

• Instruction-tuning language models based on the 10,378 dialogue pairs.
• Training models to generate natural responses based on Romanized South Asian speech patterns.
• Developing conversational AI for code-mixed Hindi-English (Hinglish) domains.
• Studying domain-diverse dialogue structures in the described base and expansion subsets.

Strengths

• Contains 10,378 high-quality instruction-following dialogue pairs.
• Includes a domain expansion subset of 379 domain-specific examples.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the main data is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Sujalvc on Hugging Face.

Freshness

Last updated 2026-05-19 09:33:44; freshness should be verified.