Sign in to view source links and access this dataset
Description
Indic Instruct Data V0.1 aggregates seven popular instruction datasets, including subsets of Flan v2, Dolly, and OpenAssistant. The collection spans English and Hindi languages, with English subsets translated using IndicTrans2. Author ai4bharat provides chrF++ scores for quality assessment of the translations.
Use Cases
Training multilingual instruction-following models based on aggregated datasets like Flan v2 and Dolly.
Evaluating machine translation quality for Hindi-English tasks based on provided chrF++ scores.
Benchmarking instruction dataset performance across languages based on the collection of seven sources.
Developing NLP applications for Hindi language based on translated subsets of popular English datasets.
Strengths
Aggregates seven distinct instruction datasets, including Flan v2 (67k sample subset), Dolly, and Anthropic-HHH (5k sample subset).
Includes translated Hindi versions of English subsets using IndicTrans2.
Provides chrF++ scores for quality assessment of the translations.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Last updated 2024-03-01 04:07:46; freshness should be verified.
Provenance
Source
ai4bharat
Collection Method
Collection and translation of existing instruction datasets.
Freshness
Last updated 2024-03-01 04:07:46.
License is unknown; restrictions should be verified before use.