Small language models, Indic-language tokenizers, continued pretraining, Telugu
Qwen3-0.6B running fully in your browser via WebGPU.