AksharaLP-MultiIndic-8M — Sourojit Dhua, 2026
https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M

Combined model-card licence: CC-BY-NC-4.0. NON-COMMERCIAL RELEASE. ATTRIBUTION REQUIRED.

ATTRIBUTION REQUIRED

Any use of this model, the demo, a derivative, or a redistribution — including research and demos — must credit Sourojit Dhua (sourojitdotcom) and link to the model page:
https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M

CC-BY-NC-4.0 requires attribution and forbids commercial use.

Ready-to-copy credit line:

    AksharaLP-MultiIndic-8M by Sourojit Dhua (sourojitdotcom) — https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M

BibTeX:

    @misc{dhua_aksharalp_multiindic_8m_2026,
      author = {Sourojit Dhua},
      title  = {AksharaLP-MultiIndic-8M},
      year   = {2026},
      url    = {https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M}
    }

The new Bengali/Telugu/Kannada/Tamil weights include Samanantar-derived Aksharantar rows. The upstream Samanantar card specifies CC-BY-NC-4.0; other publisher pages describe CC0 packaging. This release does not certify those weights as commercially clean. Preserve source attributions and applicable upstream terms. Selecting CC-BY-NC does not resolve all derivative-rights questions.

Original Hindi word/sentence components and runtime code retain Apache-2.0 and their existing notices. They are independently available at:
https://huggingface.co/sourojitdotcom/AksharaLP-1M-v2
https://huggingface.co/sourojitdotcom/AksharaLP-Sentence-2.8M

Attributions:
Aksharantar — AI4Bharat, Madhani et al. (2022), CC0 mined/existing packaging and CC-BY manual data.
Samanantar — AI4Bharat, Ramesh et al. (2022), upstream card CC-BY-NC-4.0.
Dakshina — Google Research, Roark et al., Processing South Asian Languages Written in the Latin Script, LREC 2020, https://github.com/google-research-datasets/dakshina ; CC-BY-SA-4.0 https://creativecommons.org/licenses/by-sa/4.0/ . Source-attributed training pairs appear inside Aksharantar. Retain the upstream terms.
IndicCorp v2 — AI4Bharat, Doddapaneni et al. (ACL 2023), CC0. Dictionaries and Hindi LM training.
IndicXlit — AI4Bharat, MIT. Distillation teacher and comparison model, not redistributed.
Inter and Noto Sans Indic fonts in the Space — SIL Open Font License 1.1 (included beside each font).

For the unchanged Hindi components, retain the following original notice as well:

AksharaLP-Sentence-2.8M by Sourojit Dhua
https://huggingface.co/sourojitdotcom/AksharaLP-Sentence-2.8M
Copyright 2026 Sourojit Dhua

AksharaLP-Sentence-2.8M is licensed under the Apache License, Version 2.0 (see LICENSE).
Commercial use is permitted.

What is in here
  * The v2 word model: a 989K-parameter character Transformer, its reranker, the dictionary pack
    (lexicon + word-bigram language model) and the JavaScript/Python runtime, unchanged from
    AksharaLP-1M-v2 (https://huggingface.co/sourojitdotcom/AksharaLP-1M-v2).
  * The sentence add-on: nlm.q8 (a 2,770,752-parameter Hindi neural language model trained only on
    IndicCorp v2 Hindi, CC0, with our own tokenizer, training code and runtime), plus nlm.mjs, v3.mjs
    and sent.mjs. This add-on and its training data are commercially clean; the full system still
    inherits the v2 word package's terms, which are Apache-2.0 as set out here.

ATTRIBUTION
Under Section 4(d) of the Apache License, any redistribution of this work or of a Derivative Work
(for example: shipping these files in an app, a web page, a browser extension or an SDK, or
releasing a fine-tuned, quantized or converted model based on it) must include a readable copy of
this NOTICE file. Please use the following attribution text:

    Transliteration by AksharaLP (Sourojit Dhua) — https://huggingface.co/sourojitdotcom/AksharaLP-Sentence-2.8M

Recommended in-app credit (e.g. in an About, Settings or Licenses screen):

    Hindi typing powered by AksharaLP by Sourojit Dhua

The model files in this repository were built with the help of the following
third-party resources. No third-party data is redistributed in this repository
other than as learned model parameters and derived word lists.

------------------------------------------------------------------------------
IndicCorp v2 — AI4Bharat
  https://github.com/AI4Bharat/IndicBERT/tree/main#indiccorp-v2
  License: CC0
  Used for: training the sentence language model (nlm.q8); the word-model lexicon and
  word-bigram language model.

Aksharantar — AI4Bharat
  https://huggingface.co/datasets/ai4bharat/Aksharantar
  License: CC0 (mined data and existing sources) / CC-BY (manually collected data)
  Used for: romanized <-> Devanagari word pairs for training the word model.

IndicXlit — AI4Bharat
  https://github.com/AI4Bharat/IndicXlit
  License: MIT
  Used for: teacher model for word-model distillation.

CMUdict — Carnegie Mellon University
  https://github.com/cmusphinx/cmudict
  License: BSD-2-Clause-style
  Used for: English word cues in the word model.

Fonts served by the demo Space (Inter, Noto Sans Devanagari) are under the SIL Open Font License 1.1.