Closed vyaslkv closed 3 years ago
first one is google tanslation of the middle hindi text and right one is the result of your work (pipe separated) Google result | original hindi text | Your results
one thing I can do is remove the back tick part and put that back to the translated text
All these examples are of scientific sentences. I don't think I added such data in training. I added sentence-pairs mostly from open datasets + wiki + news custom datasets.
Google updates model frequently ... so only option to improve models from this repo is adding new data + training new model at least each month
. Data is the key factor.
ok got it Swapnil Thanks!!