An indic-multilingual small language model (~150M parameters) capable of understanding English, Hindi and Mizo. It is based on the DeepSeek-V3 architecture and uses a byte-level BPE tokeniser.
Files info
Base model