MuRIL
A multilingual text encoder for Indian languages released by Google Research, based on the BERT architecture and specifically optimized for Indian-language vocabulary.
Timeline (last 90 days)
A study based on a corpus containing 131,000 sentences covering 11 Indian scripts compared the tokenizer performance of models such as TinyLlama-1.1B, mBERT, and MuRIL.
MuRIL is an encoder with an Indian-language vocabulary, and its parameter count is only one-fifth of TinyLlama's.
MuRIL achieves a reading efficiency of 1.23-2.26 tokens/word on all 11 tested Indian scripts, with zero UNK and zero byte fallback.
MuRIL is an encoder model pretrained by Google specifically for Indian languages based on the BERT architecture; its vocabulary incorporates a large number of subword units from Indian scripts, and it uses transliteration- and translation-based training objectives to enhance cross-script alignment.
All Facts (4)
A study based on a corpus containing 131,000 sentences covering 11 Indian scripts compared the tokenizer performance of models such as TinyLlama-1.1B, mBERT, and MuRIL.
50%UnverifiedMuRIL is an encoder with an Indian-language vocabulary, and its parameter count is only one-fifth of TinyLlama's.
50%UnverifiedMuRIL achieves a reading efficiency of 1.23-2.26 tokens/word on all 11 tested Indian scripts, with zero UNK and zero byte fallback.
50%UnverifiedMuRIL is an encoder model pretrained by Google specifically for Indian languages based on the BERT architecture; its vocabulary incorporates a large number of subword units from Indian scripts, and it uses transliteration- and translation-based training objectives to enhance cross-script alignment.
50%