ThamizhiPOSt Version 1.0 has been trained using 220K tagged tokens. This provides the best results for Tamil POS tagging among available neural based POS taggers.
We will be improving this by integrating a morphological analyser in the version 2.0.
Trained models can be found here: http://nlp-tools.uom.lk/thamizhi-pos/models/models and the instructions to use can be found in the readme given here: https://github.com/sarves/thamizhi-pos