normalization in hindi

[godkillok]

HI
I found the pre-train model have bad result on normalized hindi.
eg.
print(identifier.rank("तुम कहाँ जा रहे हो")) # this on is correct
[('hi', 0.5811032824612302), ('ne', 0.41881502578401597), ('mr', 8.169175475378807e-05)....]

print(identifier.rank("tum kahaan ja rahe ho"))
[('fi', 0.8528903875811155), ('et', 0.130601343501141).......]
and there most sentences like "tum kahaan ja rahe ho" is not correct.
Any idea?