AI4BharatIndicTrans

IndicTrans First-Generation Releases

The original IndicTrans project trained Transformer-4x translation models on Samanantar. It lists English-to-Indic, Indic-to-English and Indic-to-Indic v0.3 downloads.

Access not confirmed

This earlier family translated sentences

Its training limit is 200 subword tokens. The command-line workflow truncates longer sentences, while the paragraph helper splits input. The project recommends IndicTrans2. The v0.3 identifiers are retained here as historical releases without claiming a working current download.Original IndicTrans releases and sentence-length limits

Do not send a long document through the old sentence limit

Practical guidance. A fluent output can still omit text that was truncated before translation. Prefer an accessible model with a documented length appropriate for the source material.

Specifications & access

IndicTrans First-Generation Releases specifications
DetailValue and conditions
StatusAccess not confirmed
Model identifiersIndicTrans v0.3 English-to-IndicIndicTrans v0.3 Indic-to-EnglishIndicTrans v0.3 Indic-to-Indic
DesignApproximately 434M parameters; eleven Indian languages with shared Devanagari processing.Original IndicTrans releases and sentence-length limits
TermsMIT.Original IndicTrans releases and sentence-length limits
AccessAll three primary archive routes returned access errors during this review. Availability elsewhere was not confirmed.Original IndicTrans releases and sentence-length limits

Related models

  • 1.1-billion-parameter translation model from English into 22 Indian languages.

    Launch date not confirmedRestricted accessDownloadable
  • Billion-parameter-class IndicTrans2 model for translation from 22 Indian languages into English.

    Launch date not confirmedRestricted accessDownloadable
  • Billion-parameter-class IndicTrans2 model for translation between Indian languages.

    Launch date not confirmedRestricted accessDownloadable