Research
Full technical reports for the models we release: architecture, training at a high level, evaluation with every setting stated, and limitations. Each report is also available as a PDF.
Technical report ·
Gargi: a 384M thinking language model trained from scratch on about 17B tokens
Technical report for Gargi and Gargi-Base, 384M-parameter models trained from scratch on about 17B tokens, with an optional thinking mode and a 32K context: architecture, training at a high level, inference, evaluation with every setting stated, long-context recall, quantisation and limitations.
Technical report ·
Lilavati: Pocket Mathematician
Technical report for Lilavati, a 95M-parameter model trained from scratch that solves maths word problems by writing Python, and Lilavati-Checker: architecture, inference, evaluation with every setting stated, robustness, contamination audit and limitations.
Technical report ·
Chitra Tiny: a 212M-parameter text-to-image model trained from scratch for about $100
Technical report for Chitra Tiny, a 212M-parameter text-to-image model trained from scratch on public-domain and CC0 images: architecture, training at a high level, the full compute bill, FID, CLIP and GenEval results, speed and limitations.
Technical report ·
Udgam-384M: a small on-device tool-calling model trained from scratch
Technical report for Udgam-384M-Base, Base-4K and Chat, 384M-parameter models trained from scratch for tool calling on the device: architecture, inference, evaluation against other small open models with every setting stated, quantisation, safety and limitations.