MT for Hindi to English is not converging

Hi,
I am doing MT for Hindi to English. I have little less than 1 million sentences(9,00,000 approx.). But model is not doing good at all.
I trained it for 500000 steps.
I used following command
python train.py -data data/demo -save_model demo-model -gpuid 0 -gpu_ranks 0 -train_steps 500000 -encoder_type brnn -layers 4 -rnn_size 800

The prediction output I am getting is following, where not a single English word is coming out. But for some sentences I am getting English words, but they are not relevant at all.

SENT 1: (‘कॉल’, ‘हेस’, ‘बीन’, ‘पुट’, ‘ऑन’, ‘होल्ड’, ‘प्लीज’, ‘स्टे’, ‘ऑनलाइन’, ‘आप’, ‘का’, ‘कॉल’, ‘होल्ड’, ‘पर’, ‘ही’, ‘कृपया’, ‘प्रतीक्षा’, ‘कीजिये’, ‘योर’, ‘कॉल’, ‘हेस’, ‘बीन’, ‘पुट’, ‘ऑन’, ‘होल्ड’, ‘प्लीज’, ‘स्टे’, ‘ऑन’, ‘ध’, ‘लाइन’, ‘आप’, ‘का’, ‘कॉल’, ‘होल्ड’, ‘पर’, ‘ही’, ‘कृपया’, ‘प्रतीक्षा’, ‘कीजिये’, ‘योर’, ‘कॉल’, ‘हेस’, ‘बीन’, ‘पुट’, ‘ऑन’, ‘होल्ड’, ‘प्लीज’, ‘स्टे’, ‘ऑन’, ‘ध’, ‘लाइन’)
PRED 1: आप
PRED SCORE: -3.3478
PRED AVG SCORE: -3.3478, PRED PPL: 28.4397

Apart from parameters mentioned in above command, I am using default ones.

So please guide me where I am doing wrong.

can you please tell where did you find the dataset(1M parallel corpus) for english to hindi…we are working on the same,we need more data…Hope you respond!

I think this database was in-house prepared and cleaned after crawling the data from net.

1 Like

if possible can you share with me?

it will be a great help for us