Research.
ByteZira explores AI experimentation, engineering and technical learning. We report what we have measured and label what we have not.
Featured: ByteZira Vaani-350M
Training scale, measured performance and estimated capability are kept separate here. Only the first is reported so far.
- Architecture
- Custom decoder-only Transformer
- Parameters
- Approximately 350 million
- Training tokens
- Around 3.3 billion (reported, pending verification)
- Framework and tooling
- To be published
- Intended use
- Research and experimentation
- Known limitations
- To be published with model documentation
- Evaluation
- No results published yet
- Status
- Experimental research model
Research methodology
We aim for reproducible experiments, documented evaluation methods and responsible reporting, including limitations.
Publications and technical notes
No notes have been published yet. They will appear here.
Future research
Exploratory and planned only. None of this is completed work.