metadata
license: afl-3.0
datasets:
- WillHeld/hinglish_top
language:
- en
- hi
metrics:
- accuracy
library_name: transformers
pipeline_tag: fill-mask
SRDberta
This is a BERT model trained for Masked Language Modeling for Hinglish Data.
Hinglish is a term used to describe the hybrid language spoken in India, which combines elements of Hindi and English. It is commonly used in informal conversations and in media such as Bollywood films
Dataset
Hinglish-Top Dataset columns
- en_query
- cs_query
- en_parse
- cs_parse
- domain
Training
| Epochs | Train Loss |
|---|---|
| 4th | 0.251 |
The model was trained only for 4 epochs due to the GPU limitations. The model will give far better results with 10 epochs
Inference
from transformers import AutoTokenizer, AutoModelForMaskedLM, pipeline
tokenizer = AutoTokenizer.from_pretrained("SRDdev/SRDBerta")
model = AutoModelForMaskedLM.from_pretrained("SRDdev/SRDBerta")
fill = pipeline('fill-mask', model='SRDberta', tokenizer='SRDberta')
fill_mask = fill.tokenizer.mask_token
fill(f'Aap {fill_mask} ho?')
Citation
Author: @SRDdev
Name : Shreyas Dixit
framework : Pytorch
Year: Jan 2023
Pipeline : fill-mask
Github : https://github.com/SRDdev
LinkedIn : https://www.linkedin.com/in/srddev/