James smith's picture

James smith

goodluck-03301
ยท

AI & ML interests

None yet

Recent Activity

repliedto kalyan-ks's post about 1 hour ago
NandiBERT >> IndicBERTv2, ModernBERT and Ettin Encoder NandiBERT - ๐Ž๐ฉ๐ž๐ง ๐’๐Ž๐“๐€ ๐„๐ง๐œ๐จ๐๐ž๐ซ ๐Œ๐จ๐๐ž๐ฅ ๐Ÿ๐จ๐ซ ๐ˆ๐ง๐๐ข๐š๐ง ๐‹๐š๐ง๐ ๐ฎ๐š๐ ๐ž๐ฌ NandiBERT, SOTA Encoder model for Indian languages which is - SOTA for Indic languages outperforming IndicBERT v2 - among the best for English language outperforming Ettin Encoder and ModernBERT models. ๐Š๐ž๐ฒ ๐…๐ž๐š๐ญ๐ฎ๐ซ๐ž๐ฌ ๐จ๐Ÿ ๐๐š๐ง๐๐ข๐๐„๐‘๐“ [1] Multilingual Encoder : A single model supporting English and 11 Indian languages like Hindi, Telugu, Tamil etc. [2] Two Model Sizes : Available in two sizes namely NandiBERT-150M and NandiBERT-450M [3] Bidirectional Transformer Architecture: Uses bidirectional attention so that every token can attend to every other input token [4] Long Context: NandiBERT-150M supports 16K context and NandiBERT-450M supports 32K context [5] Large Vocabulary: Uses 100,352 vocabulary shared across languages [6] SOTA Performance: Outperforms IndicBERT v2 (for Indic languages) and ModernBERT, Ettin Encoder (for English language) [7] Open Weights and License: Models weights are available on HF platform. NandiBERT-150M: https://huggingface.co/FrontiersMind/NandiBERT-150M NandiBERT-450M: https://huggingface.co/FrontiersMind/NandiBERT-450M
View all activity

Organizations

None yet