| language: | |
| - multilingual | |
| language_details: "NLLB-200 supports 200 languages (see facebook/nllb-200-distilled-600M)" | |
| license: "cc-by-nc-4.0" | |
| tags: | |
| - translation | |
| - coreml | |
| - nllb | |
| - apple | |
| - ios | |
| - mlpackage | |
| - float32 | |
| pipeline_tag: translation | |
| datasets: | |
| - flores-200 | |
| base_model: facebook/nllb-200-distilled-600M | |
| # NLLB-200 Core ML 1024 — FLOAT32 (private snapshot) | |
| **Private** Hub repo: full **float32** Core ML `mlpackage` tree (encoder + KV decoder init/step), ~**5 GB** on disk. Same layout as local `output/nllb200-coreml-1024`. | |
| This is a **separate repository** from: | |
| - Public or other `nllb200-coreml-1024` variants on the Hub | |
| - [nllb200-coreml-1024-pal8](https://huggingface.co/aoiandroid/nllb200-coreml-1024-pal8) (8-bit weight-only palettized, ~1.3 GB) | |
| Use **float32** when you need maximum numerical stability; use **pal8** for smaller download and faster inference on tested Mac setups. | |
| ## Base model (official) | |
| Derived from **[facebook/nllb-200-distilled-600M](https://huggingface.co/facebook/nllb-200-distilled-600M)** — same license and use restrictions as the base model card ([CC-BY-NC-4.0](https://creativecommons.org/licenses/by-nc/4.0/)). | |
| | Item | Value | | |
| |------|--------| | |
| | Max length | 1024 | | |
| | Packages | `NLLB_Encoder_1024`, `NLLB_Decoder_1024_init`, `NLLB_Decoder_1024_step` + `tokenizer/` + `config.json` | | |
| ## App / model ID | |
| on-device iOS client: **`nllb200-coreml-1024`** with `NLLBCoreMLTranslationService` (same file names as this repo). | |
| ## License and citation | |
| Same as [README_nllb200_coreml_1024.md](local iOS project (private)) and the NLLB paper — cite `facebook/nllb-200-distilled-600M` and NLLB-200. | |