Government Of India
A-
A
A+
KhasiBERT
Khasi language model trained on 3.6M sentences using RoBERTa architecture. 110M parameters. Supports NLP tasks for Khasi text processing.
About Model
KhasiBERT is the first transformer-based language model developed specifically for the Khasi language of Meghalaya. This foundational model addresses the computational linguistics needs for Meghalaya's 1.4+ million Khasi speakers and enables digital language processing infrastructure for the state.
The model implements the RoBERTa architecture with 12 transformer layers, 768 hidden dimensions, and 12 attention heads, totaling 110,652,416 parameters. KhasiBERT was trained using masked language modeling on a corpus of 3,621,116 Khasi sentences collected and preprocessed to represent diverse linguistic patterns of the language.
A custom Byte-Level BPE tokenizer with 32,000 vocabulary tokens was developed specifically for the Khasi language to handle its Austroasiatic linguistic characteristics effectively. Training utilized mixed-precision (FP16) optimization with a batch size of 24, learning rate of 5e-5, and AdamW optimizer over 150,880 training steps.
The model supports standard transformer operations including masked token prediction, contextualized embeddings generation, and can be fine-tuned for downstream tasks such as text classification, sentiment analysis, named entity recognition, and question answering in Khasi. Input sequences are limited to 512 tokens with standard special tokens for beginning-of-sequence, end-of-sequence, padding, unknown tokens, and masking.
Technical specifications include GELU activation functions, 0.1 dropout probability, layer normalization with epsilon 1e-12, and positional embeddings up to 514 positions. The model weights are distributed in Safetensors format and are compatible with the Transformers library ecosystem for integration with existing NLP workflows.
KhasiBERT
Metadata
Creative Commons Attribution Non Commercial 4.0
MWirelabs
Text Generation
PyTorch
Open
Education and Skill Development
04/09/25 02:55:59
0
Activity Overview
1
27 
0 
853
Tags
- Bert
- safetensors
- endpoints_compatible
- autotrain_compatible
- Fill-Mask
- Indian Language
- low-resource
- region:us
- kha
- khasi
- Meghalaya
- austroasiatic
- masked-lm
- roberta
- foundational-model
- digital-india
License Control
Creative Commons Attribution Non Commercial 4.0
Related Models
Kren-M
Northeast India's first AI language model. Kren-M is a 2.6B parameter bilingual model for Khasi-English, built on Gemma-2-2B. Features Kren-NE custom tokenizer covering 7 NE languages (Khasi, Garo, Mizo, Assamese, Manipuri, Nagamese, Nyishi) with 35.7% efficiency gain. Trained on 5.43M Khasi sentences. Capabilities: bidirectional translation, natural conversation, cultural context. Designed for language preservation across Northeast India
Tokenizer
low-resource
Garo
northeast-india
khasi
Northeast India
continued-pretraining
bilingual
Kren-M
Northeast India Languages
Foundational model
Indian Languages
Instruction-Tuning
- 0
- 30
- 0
745
Updated 5 month(s) ago
MWIRE LABS
Kren v1: Khasi Generative Language Model
Kren v1 is the first Khasi generative language model, trained on 1M lines, pioneering encoder-to-decoder adaptation for low-resource AI.
Low-Resource NLP
khasi
khasi-culture
Meghalaya
Indigenous Language
Northeast India
Encoder-to-Decoder
AI for Culture
MWire Labs
Natural Language Processing
- 0
0 - 390.67 MB
- 8
Updated 7 month(s) ago
MWIRE LABS
More Models from MWire Labs
Mizo OCR - Text Recognition for Mizo Language
OCR model for the Mizo language achieving 90.68% character accuracy on synthetic and curated printed text
Mizo
northeast-india
trocr
low-resource
OCR
Image-to-Text
- 0
- 2
- 0
- 162
Updated 2 month(s) ago
MWIRE LABS
NE-OCR
NE-OCR is a multilingual Optical Character Recognition model developed by MWire Labs to accurately recognize printed text from documents in Northeast Indian languages. The model supports Assamese, Bodo, English, Garo, Hindi, Khasi, Kokborok, Meitei (Bengali script), Meitei (Meitei Mayek script), Mizo, Nagamese, and Nyishi. It is designed to enable reliable digitization of books, newspapers, government records, educational materials, and cultural archives from Northeast India where mainstream OCR
Printed Text Recognition
Northeast India OCR
Multilingual OCR
vitstr
doctr
BODO
Optical Character Recognition
Mizo
khasi
northeast-india
Nagamese
Meitei
Nyishi
Kokborok
Garo
OCR
- 0
- 7
- 0
- 124
Updated 2 month(s) ago
MWIRE LABS
Nagamese Speech-to-Text
Automatic Speech Recognition (ASR) model for Nagamese speech, designed to transcribe spoken Nagamese into text for real-world usage.
whisper
Speech Recognition
ASR
Automatic Speech Recognition
low-resource-language
Nagamese
- 0
- 3
- 0
- 117
Updated 2 month(s) ago
MWIRE LABS
Garo OCR - Text Recognition for Garo
OCR model for the Garo language achieving 93.13% character accuracy.
Garo
florence-2
northeast-india
OCR
Image-to-Text
- 0
- 0
- 0
- 153
Updated 2 month(s) ago
MWIRE LABS
Northeast Language Identification
NE-LID is a fast and accurate language identification model for Northeast Indian languages using character level features. It is designed for low resource and script diverse text and achieves high accuracy on short sentences.
Multilingual
language identification
low-resource
northeast-india
MWire Labs
fasttext
fastText
- 1
- 11
- 0
- 588
Updated 4 month(s) ago
MWIRE LABS
NortheastNER
NortheastNER is a token classification model built on XLM-RoBERTa and fine-tuned on ~25k sentences from gazetteers, news, and cultural texts across Northeast India. It detects region-specific entities, places, tribes, festivals, tourist sites, flora, fauna, and experimental local names; ideal for low-resource NER, regional search, cultural analytics, and knowledge graph applications.
Conservation
Token Classification
NER
XLM-RoBERTa
low-resource
northeast-india
Meghalaya
Northeast India
- 0
- 12
- 0
- 304
Updated 5 month(s) ago
MWIRE LABS
Kren-M
Northeast India's first AI language model. Kren-M is a 2.6B parameter bilingual model for Khasi-English, built on Gemma-2-2B. Features Kren-NE custom tokenizer covering 7 NE languages (Khasi, Garo, Mizo, Assamese, Manipuri, Nagamese, Nyishi) with 35.7% efficiency gain. Trained on 5.43M Khasi sentences. Capabilities: bidirectional translation, natural conversation, cultural context. Designed for language preservation across Northeast India
Northeast India Languages
Tokenizer
Foundational model
Kren-M
bilingual
continued-pretraining
Northeast India
khasi
northeast-india
Garo
low-resource
Instruction-Tuning
Indian Languages
- 0
- 30
- 0
- 745
Updated 5 month(s) ago
MWIRE LABS
NE-BERT
NE-BERT is Northeast India's first domain-specific multilingual foundation model. Built on the ModernBERT architecture and trained on 8.3 million sentences, it supports 9 regional languages: Assamese, Khasi, Garo, Manipuri (Meitei), Mizo, Nyishi, Nagamese, Kokborok, and Pnar. It achieves State-of-the-Art performance on regional benchmarks and offers 1.6x faster inference, bridging the digital divide for low-resource languages.
kokborok
Mizo
token-efficiency
northeast bert
low-resource-NLP
Masked Language Modeling
Assamese
Garo
Nyishi
Meitei
Nagamese
northeast-india
khasi
A'chik
modernbert
mwirelabs
Pnar
- 0
- 20
- 0
- 550
Updated 5 month(s) ago
MWIRE LABS
KhasiBERT
Khasi language model trained on 3.6M sentences using RoBERTa architecture. 110M parameters. Supports NLP tasks for Khasi text processing.
austroasiatic
roberta
foundational-model
digital-india
Bert
safetensors
endpoints_compatible
autotrain_compatible
Fill-Mask
Indian Language
low-resource
region:us
kha
khasi
Meghalaya
masked-lm
- 1
- 27
- 0
- 854
Updated 8 month(s) ago
MWIRE LABS
Khasi English Semantic Search Model
Khasi-English semantic search model, trained on 66,794 pairs with 0.69-0.74 similarity. ~90MB, supports Meghalaya tourism/culture. By MWirelabs
Meghalaya
safetensors
Sentence Similarity
cross-lingual
en
autotrain_compatible
semantic search
sentence-transformers
license:cc0-1.0
kha
khasi
text-embeddings-inference
khasi-culture
- 0
- 24
- 0
- 775
Updated 8 month(s) ago
MWIRE LABS
© 2026 - Copyright AIKosh. All rights reserved.