sourojitdotcom commited on
Commit
37a22c3
·
verified ·
1 Parent(s): 5e1077b

Public release: attribution required under CC-BY-NC-4.0; drop planned-retrain wording

Browse files
Files changed (6) hide show
  1. ATTRIBUTION +19 -0
  2. NOTICE +23 -2
  3. README.md +33 -8
  4. SHA256SUMS +5 -4
  5. assets/banner.png +2 -2
  6. assets/banner.svg +1 -1
ATTRIBUTION ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ATTRIBUTION REQUIRED
2
+
3
+ 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:
4
+ https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M
5
+
6
+ CC-BY-NC-4.0 requires attribution and forbids commercial use.
7
+
8
+ Ready-to-copy credit line:
9
+
10
+ AksharaLP-MultiIndic-8M by Sourojit Dhua (sourojitdotcom) — https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M
11
+
12
+ BibTeX:
13
+
14
+ @misc{dhua_aksharalp_multiindic_8m_2026,
15
+ author = {Sourojit Dhua},
16
+ title = {AksharaLP-MultiIndic-8M},
17
+ year = {2026},
18
+ url = {https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M}
19
+ }
NOTICE CHANGED
@@ -1,8 +1,29 @@
1
  AksharaLP-MultiIndic-8M — Sourojit Dhua, 2026
2
  https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M
3
 
4
- Combined model-card licence: CC-BY-NC-4.0. NON-COMMERCIAL RELEASE.
5
- 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. An Apache-2.0 clean-data retrain is planned. Preserve source attributions and applicable upstream terms. Selecting CC-BY-NC does not resolve all derivative-rights questions.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
 
7
  Original Hindi word/sentence components and runtime code retain Apache-2.0 and their existing notices. They are independently available at:
8
  https://huggingface.co/sourojitdotcom/AksharaLP-1M-v2
 
1
  AksharaLP-MultiIndic-8M — Sourojit Dhua, 2026
2
  https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M
3
 
4
+ Combined model-card licence: CC-BY-NC-4.0. NON-COMMERCIAL RELEASE. ATTRIBUTION REQUIRED.
5
+
6
+ ATTRIBUTION REQUIRED
7
+
8
+ 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:
9
+ https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M
10
+
11
+ CC-BY-NC-4.0 requires attribution and forbids commercial use.
12
+
13
+ Ready-to-copy credit line:
14
+
15
+ AksharaLP-MultiIndic-8M by Sourojit Dhua (sourojitdotcom) — https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M
16
+
17
+ BibTeX:
18
+
19
+ @misc{dhua_aksharalp_multiindic_8m_2026,
20
+ author = {Sourojit Dhua},
21
+ title = {AksharaLP-MultiIndic-8M},
22
+ year = {2026},
23
+ url = {https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M}
24
+ }
25
+
26
+ 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.
27
 
28
  Original Hindi word/sentence components and runtime code retain Apache-2.0 and their existing notices. They are independently available at:
29
  https://huggingface.co/sourojitdotcom/AksharaLP-1M-v2
README.md CHANGED
@@ -25,7 +25,7 @@ metrics:
25
  ---
26
 
27
  <p align="center"><img src="assets/banner.svg" alt="AksharaLP MultiIndic — five scripts, one small footprint" width="100%"/></p>
28
- <p align="center"><a href="https://huggingface.co/spaces/sourojitdotcom/AksharaLP-MultiIndic-demo">Try the private browser demo ↗</a> · <a href="#benchmarks">Measured benchmarks</a> · <a href="#gguf-containers">13 GGUF files</a> · <a href="#licence--data">Licence & data</a></p>
29
 
30
  # AksharaLP-MultiIndic-8M
31
 
@@ -33,7 +33,7 @@ metrics:
33
 
34
  **Hindi · Bengali · Telugu · Kannada · Tamil** — familiar Roman letters in, native script out. Five independent ~989K-parameter character Transformers, a lazily loaded dictionary per language, and the already-shipped **2,770,752-parameter Hindi sentence LM**. No inference server and no GPU required for inference. This is **transliteration, not translation or a general-purpose language model**.
35
 
36
- > **Private review · non-commercial release.** This bundle's model-card licence is **CC-BY-NC-4.0**, not Apache-2.0. The new Bengali/Telugu/Kannada/Tamil weights include Samanantar-derived Aksharantar rows with non-commercial upstream terms. A commercially clean Apache-2.0 retrain is planned. The Hindi word and sentence components remain independently available under Apache-2.0 in their original repositories. See [Licence & data](#licence--data).
37
 
38
  <table><tr><td align="center"><h2>4.94M</h2>five word models combined</td><td align="center"><h2>7.72M</h2>with the optional Hindi LM</td><td align="center"><h2>20.55 MB</h2>all word payloads, gzip-9</td><td align="center"><h2>23.48 MB</h2>including Hindi sentence mode</td></tr></table>
39
 
@@ -197,14 +197,14 @@ Browser gzip decoding requires `DecompressionStream` support; Node uses its buil
197
 
198
  ## Python reference
199
 
200
- Requires `numpy` and `torch`. For a private review repository, authenticate with Hugging Face normally; never put a write token in browser code.
201
 
202
  ```python
203
  from pathlib import Path
204
  import sys
205
  from huggingface_hub import snapshot_download
206
 
207
- root = Path(snapshot_download('sourojitdotcom/AksharaLP-MultiIndic-8M', token=True))
208
  sys.path.insert(0, str(root / 'python'))
209
  from multiindic import MultiIndic
210
 
@@ -264,9 +264,34 @@ engine.load_sentence_gguf(root / 'gguf/aksharalp-hi-sentence-q8.gguf')
264
  print(engine.transliterate('main theek hoon'))
265
  ```
266
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
267
  ## Licence & data
268
 
269
- **Model-card licence: CC-BY-NC-4.0. This release is non-commercial.** The Bengali, Telugu, Kannada and Tamil weights contain `aks-Samanantar` Aksharantar training rows. The [current upstream Samanantar card](https://huggingface.co/datasets/ai4bharat/samanantar) states **CC-BY-NC-4.0**. Aksharantar and the publisher's Samanantar packaging page describe CC0 packaging, so provenance terms conflict; this release does **not** certify those weights as commercially clean. A commercially clean **Apache-2.0 retrain is planned**, and would be separately evaluated before replacement.
270
 
271
  | Resource | Terms / role |
272
  |---|---|
@@ -289,7 +314,7 @@ The **Hindi components are also separately available under Apache-2.0**: [Akshar
289
 
290
  **Does every language understand a sentence?** No. Only Hindi has neural sentence rescoring. Others concatenate independent word predictions, preserving text layout. Hindi rescoring cannot recover a spelling outside its candidate list.
291
 
292
- **Can I use it commercially?** Not this combined release. Its model-card licence is CC-BY-NC-4.0. Use independently Apache-licensed Hindi releases where appropriate, or wait for a cleared retrain. Retain all source attributions and review applicable component terms.
293
 
294
  **Does text leave the device?** The demo fetches local model assets from the Space when loading a language. Typed text is processed in the browser and is not sent to an inference API or telemetry service. Once the selected assets are loaded, typing needs no network. This is not a promise to avoid the initial asset download.
295
 
@@ -299,7 +324,7 @@ The **Hindi components are also separately available under Apache-2.0**: [Akshar
299
 
300
  Names, ambiguous short words, unseen vocabulary, English code-switching, dialects and chat abbreviations can fail. The output must be user-reviewable; do not silently replace important legal, medical or financial text. Candidate suggestions are uncalibrated. Input encoding truncates a word after 31 Roman characters. Type-as-you-go inference is debounced; older phones and long sentences may be slow. All measured compute here is CPU, not an on-device mobile benchmark. Tiny low-bit downloads still expand to float32 arrays and large dictionary structures in RAM.
301
 
302
- This private release records the current artifacts. No “best in class,” general-LLM capability, sub-5-MB complete bundle, commercial-clean, or multilingual-sentence claim is made.
303
 
304
  ---
305
- **AksharaLP · Sourojit Dhua · 2026-11-10 (IST)** · Private review copy. Do not make public without the owner's approval.
 
25
  ---
26
 
27
  <p align="center"><img src="assets/banner.svg" alt="AksharaLP MultiIndic — five scripts, one small footprint" width="100%"/></p>
28
+ <p align="center"><a href="https://huggingface.co/spaces/sourojitdotcom/AksharaLP-MultiIndic-demo">Try the browser demo ↗</a> · <a href="#benchmarks">Measured benchmarks</a> · <a href="#gguf-containers">13 GGUF files</a> · <a href="#attribution-required">Attribution required</a> · <a href="#licence--data">Licence & data</a></p>
29
 
30
  # AksharaLP-MultiIndic-8M
31
 
 
33
 
34
  **Hindi · Bengali · Telugu · Kannada · Tamil** — familiar Roman letters in, native script out. Five independent ~989K-parameter character Transformers, a lazily loaded dictionary per language, and the already-shipped **2,770,752-parameter Hindi sentence LM**. No inference server and no GPU required for inference. This is **transliteration, not translation or a general-purpose language model**.
35
 
36
+ > **Attribution required. Non-commercial (CC-BY-NC-4.0).** Any use — including research, demos, derivatives, and redistribution — must credit **Sourojit Dhua (sourojitdotcom)** and link to this [model page](https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M). CC-BY-NC-4.0 requires attribution and forbids commercial use. The new Bengali/Telugu/Kannada/Tamil weights include Samanantar-derived Aksharantar rows; the [upstream Samanantar card](https://huggingface.co/datasets/ai4bharat/samanantar) states CC-BY-NC-4.0, so this release does **not** certify those weights as commercially clean. The Hindi word and sentence components remain independently available under Apache-2.0 in their original repositories. See [Attribution required](#attribution-required) and [Licence & data](#licence--data).
37
 
38
  <table><tr><td align="center"><h2>4.94M</h2>five word models combined</td><td align="center"><h2>7.72M</h2>with the optional Hindi LM</td><td align="center"><h2>20.55 MB</h2>all word payloads, gzip-9</td><td align="center"><h2>23.48 MB</h2>including Hindi sentence mode</td></tr></table>
39
 
 
197
 
198
  ## Python reference
199
 
200
+ Requires `numpy` and `torch`. The repository is public; no Hugging Face token is required to download it. Never put a write token in browser code.
201
 
202
  ```python
203
  from pathlib import Path
204
  import sys
205
  from huggingface_hub import snapshot_download
206
 
207
+ root = Path(snapshot_download('sourojitdotcom/AksharaLP-MultiIndic-8M'))
208
  sys.path.insert(0, str(root / 'python'))
209
  from multiindic import MultiIndic
210
 
 
264
  print(engine.transliterate('main theek hoon'))
265
  ```
266
 
267
+ ## Attribution required
268
+
269
+ **Any use of this model or demo — including research, demos, derivatives, and redistribution — must credit Sourojit Dhua (sourojitdotcom)** and link to this model page: [https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M](https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M).
270
+
271
+ This release is licensed under [CC-BY-NC-4.0](https://creativecommons.org/licenses/by-nc/4.0/), which **requires attribution** and **forbids commercial use**.
272
+
273
+ Ready-to-copy credit line:
274
+
275
+ ```text
276
+ AksharaLP-MultiIndic-8M by Sourojit Dhua (sourojitdotcom) — https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M
277
+ ```
278
+
279
+ BibTeX:
280
+
281
+ ```bibtex
282
+ @misc{dhua_aksharalp_multiindic_8m_2026,
283
+ author = {Sourojit Dhua},
284
+ title = {AksharaLP-MultiIndic-8M},
285
+ year = {2026},
286
+ url = {https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M}
287
+ }
288
+ ```
289
+
290
+ The same text is in [NOTICE](NOTICE) and [ATTRIBUTION](ATTRIBUTION).
291
+
292
  ## Licence & data
293
 
294
+ **Model-card licence: CC-BY-NC-4.0. This release is non-commercial, and attribution is required.** The Bengali, Telugu, Kannada and Tamil weights contain `aks-Samanantar` Aksharantar training rows. The [current upstream Samanantar card](https://huggingface.co/datasets/ai4bharat/samanantar) states **CC-BY-NC-4.0**. Aksharantar and the publisher's Samanantar packaging page describe CC0 packaging, so provenance terms conflict; this release does **not** certify those weights as commercially clean.
295
 
296
  | Resource | Terms / role |
297
  |---|---|
 
314
 
315
  **Does every language understand a sentence?** No. Only Hindi has neural sentence rescoring. Others concatenate independent word predictions, preserving text layout. Hindi rescoring cannot recover a spelling outside its candidate list.
316
 
317
+ **Can I use it commercially?** No. This combined release is CC-BY-NC-4.0: commercial use is not allowed, and any use must credit **Sourojit Dhua (sourojitdotcom)** with a link to this model page. Independently Apache-licensed Hindi releases remain available for Hindi-only use where their terms apply. Retain all source attributions and review applicable component terms.
318
 
319
  **Does text leave the device?** The demo fetches local model assets from the Space when loading a language. Typed text is processed in the browser and is not sent to an inference API or telemetry service. Once the selected assets are loaded, typing needs no network. This is not a promise to avoid the initial asset download.
320
 
 
324
 
325
  Names, ambiguous short words, unseen vocabulary, English code-switching, dialects and chat abbreviations can fail. The output must be user-reviewable; do not silently replace important legal, medical or financial text. Candidate suggestions are uncalibrated. Input encoding truncates a word after 31 Roman characters. Type-as-you-go inference is debounced; older phones and long sentences may be slow. All measured compute here is CPU, not an on-device mobile benchmark. Tiny low-bit downloads still expand to float32 arrays and large dictionary structures in RAM.
326
 
327
+ This release records the current artifacts. No “best in class,” general-LLM capability, sub-5-MB complete bundle, commercial-clean, or multilingual-sentence claim is made.
328
 
329
  ---
330
+ **AksharaLP · Sourojit Dhua (sourojitdotcom) · 2026-11-10 (IST)** · CC-BY-NC-4.0. Attribution required. Non-commercial use only.
SHA256SUMS CHANGED
@@ -1,13 +1,14 @@
1
  ef294438456f56bb1968e8725a65d841a2d0c685aa591bab663b441a860a2465 .gitattributes
 
2
  cc97a2c4ff7723a1f25179dc7c18f676fa84191bbb36ea7d0061acaf3c102559 LICENSE
3
  e61722325c93e58388069e84e208e43ca5b6294050676ee27194921cf453a071 LICENSE-APACHE-2.0
4
- 8dd4007ec7ce19774061f88c9173ef176951e329083c079728c901ce15b15f13 NOTICE
5
- edc975f244807c134d306d1376fdfb8b29073d5db540e30db0bc90371b3c1e45 README.md
6
  f6484f5caaebaf2acb68540cc2cb29402030aee22a7e45b25d37424872f769a8 aksharalp.mjs
7
  e29d4e04be4a7235d4f2b94b6578cd580486960a7068b048e1bf6c1b1799679a assets/accuracy.png
8
  c5378130af32052f6020b7cdc5a491aa5f1d11de763c37ee5491ba1eaef0bf49 assets/accuracy.svg
9
- 27da3ac213f72636663b84026093cf2a1c80f825d8d67f110466ae9df0d3d6fc assets/banner.png
10
- a9f8a61342401feb1e4bb6c1f82d0831d3fcbf35e3a6788eb7027d193b8de129 assets/banner.svg
11
  70d7aee1fe5e1e61dc7154c0aa4e33c818ff30ebbbe5d78fb5f52ba09fad1e04 assets/sizes.png
12
  e6a92bb1337fb4997c5d657e12d0b34208d0240e006aa3eb58bce6d3a9b53bec assets/sizes.svg
13
  f5689263059364546fe23824634183d8781defb4691f36d951fff63558f16afc config.json
 
1
  ef294438456f56bb1968e8725a65d841a2d0c685aa591bab663b441a860a2465 .gitattributes
2
+ fb947d2eb60cf092d963cdceb779d082ac6eae2e8ea02ce51a0dcc334f9fa4ca ATTRIBUTION
3
  cc97a2c4ff7723a1f25179dc7c18f676fa84191bbb36ea7d0061acaf3c102559 LICENSE
4
  e61722325c93e58388069e84e208e43ca5b6294050676ee27194921cf453a071 LICENSE-APACHE-2.0
5
+ 07099348c854c0a7845c83ae06d83cffdc817f40997c0ae4501a52103ea31ed7 NOTICE
6
+ 0ce210c66b634b769c9dd1a03f0cfe6f7ca79dce1f6a8d43a5c745d83e24bc07 README.md
7
  f6484f5caaebaf2acb68540cc2cb29402030aee22a7e45b25d37424872f769a8 aksharalp.mjs
8
  e29d4e04be4a7235d4f2b94b6578cd580486960a7068b048e1bf6c1b1799679a assets/accuracy.png
9
  c5378130af32052f6020b7cdc5a491aa5f1d11de763c37ee5491ba1eaef0bf49 assets/accuracy.svg
10
+ 754e34b71d892ef137be1fffdefde88df7d811a17135efc5747e1057161452d3 assets/banner.png
11
+ ef1d99cabfd5b2814b3e991a6ac9a97e2089c93e76a44d500ae01ae395e87815 assets/banner.svg
12
  70d7aee1fe5e1e61dc7154c0aa4e33c818ff30ebbbe5d78fb5f52ba09fad1e04 assets/sizes.png
13
  e6a92bb1337fb4997c5d657e12d0b34208d0240e006aa3eb58bce6d3a9b53bec assets/sizes.svg
14
  f5689263059364546fe23824634183d8781defb4691f36d951fff63558f16afc config.json
assets/banner.png CHANGED

Git LFS Details

  • SHA256: 27da3ac213f72636663b84026093cf2a1c80f825d8d67f110466ae9df0d3d6fc
  • Pointer size: 131 Bytes
  • Size of remote file: 210 kB

Git LFS Details

  • SHA256: 754e34b71d892ef137be1fffdefde88df7d811a17135efc5747e1057161452d3
  • Pointer size: 131 Bytes
  • Size of remote file: 197 kB
assets/banner.svg CHANGED