ai4bharat
IN22 Conversational
Conversational translation evaluation data from AI4Bharat. A possible starting point for language-specific evaluation after access and license review.
Hugging FaceCC BY 4.0
LOADING
LOADING
Discover open models. Understand their capabilities.
Build AI that works in the real world.
ai4bharat
Conversational translation evaluation data from AI4Bharat. A possible starting point for language-specific evaluation after access and license review.
ai4bharat
Translation evaluation dataset covering Indian languages. Use its stated license and access conditions; keep evaluation data separate from training data.
A model card tells you what a model claims. Suvyo Proof is being built to show what survives a reproducible test.
// TRUST STARTS WITH A TRACE
suite structured-outputinputs versioned + fixedvalidate schema + taskrecord revision + runtimepublish after review0 published independent evaluations.
No scores without an executed, reviewed run.
Practical, curator-authored starting points at the intersection of available models and real needs.
Find something worth buildingA support request can be syntactically valid yet route to the wrong tool. Build a compact evaluation set for route selection, ambiguity and abstention across English, Hindi and mixed-language input.
Generated SQL can look plausible while returning the wrong rows or attempting a write. Build a read-only query evaluation using synthetic schemas and expected result sets.
Document pipelines often lose the relationship between an extracted value and its source location. Prototype field extraction that retains page references and rejects unsupported values.
Why Suvyo connects discovery, reproducible evidence, contribution and practical engineering.
How to read Suvyo Proof evidence, and why a passing JSON schema is not the same as a correct answer.
A practical guide to submitting a resource, proposing a task and preserving attribution.
Independent. Indian-founded. Globally open.
Building useful infrastructure for the next wave of AI builders.
AI4Bharat
Indian-language evaluation tooling from AI4Bharat. Compare results only when task versions and evaluation settings agree.
AI4Bharat
A blueprint for creating Pretraining and Fine-Tuning datasets for Indic languages
AI4Bharat
Translation models for 22 scheduled languages of India
ai4bharat
Speech-recognition checkpoint from AI4Bharat. Real deployment needs dialect, audio-quality and domain-specific error analysis.