merged english and multilingual and updated model card

README.md

@@ -40,8 +40,7 @@ This model is ready for commercial use.
 The use of this model is governed by the [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) and the use of the post-processing scripts are licensed under [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0.txt).
 
 ### Release Date:  <br>
-Hugging Face (this repo) [nvidia/nemotron-ocr-v2-multilingual](https://huggingface.co/nvidia/nemotron-ocr-v2-multilingual) <br>
-Collection / variant hub: [nvidia/nemotron-ocr-v2](https://huggingface.co/nvidia/nemotron-ocr-v2) <br>
+Hugging Face (this repo): [nvidia/nemotron-ocr-v2](https://huggingface.co/nvidia/nemotron-ocr-v2) <br>
 Build.Nvidia.com 04/15/2026 via [https://build.nvidia.com/nvidia/nemotron-ocr-v2](https://build.nvidia.com/nvidia/nemotron-ocr-v2) <br>
 NGC 04/15/2026 via [https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo-microservices/containers/nemoretriever-ocr-v2](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo-microservices/containers/nemoretriever-ocr-v2) <br>
 
@@ -59,8 +58,8 @@ Global
 
 Nemotron OCR v2 is available in two variants:
 
-- **v2_english** — Optimized for English-language OCR with a compact recognizer for lower latency.
-- **v2_multilingual** — Supports English, Chinese (Simplified and Traditional), Japanese, Korean, and Russian with a larger recognizer to accommodate the expanded character set.
+- **v2_english** — Optimized for English-language OCR with word-level region handling.
+- **v2_multilingual** — Supports English, Chinese (Simplified and Traditional), Japanese, Korean, and Russian with line-level region handling for multilingual documents.
 
 Both variants share the same three-component architecture:
 
@@ -96,7 +95,7 @@ The two variants share an identical detector and relational architecture but dif
 | Relational model  | 2,255,419   |
 | **Total**         | **53,831,335**  |
 
-**v2_multilingual** (this repository: `checkpoints/`):
+**v2_multilingual** (from `v2_multilingual/`):
 
 | Component         | Parameters  |
 |-------------------|-------------|
@@ -218,25 +217,26 @@ Output is saved next to your input image as `<name>-annotated.<ext>` on the host
 
 3. Run the model using the following code.
 
-Use `nemotron_ocr.inference.pipeline.NemotronOCR`. With no arguments, checkpoints are downloaded from Hugging Face: **by default** the **v2 multilingual** bundle ([`nvidia/nemotron-ocr-v2-multilingual`](https://huggingface.co/nvidia/nemotron-ocr-v2-multilingual), `checkpoints/`). Use `lang="en"` for the English-optimized v2 build (`nvidia/nemotron-ocr-v2` / `v2_english/`), or pass `model_dir` to load from disk (any complete checkpoint folder; `lang` is then ignored).
+Use `nemotron_ocr.inference.pipeline_v2.NemotronOCRV2`. With no arguments, checkpoints are downloaded from Hugging Face: **by default** the **v2 multilingual** bundle (`nvidia/nemotron-ocr-v2` / `v2_multilingual/`). Use `lang="en"` for the English v2 build (`nvidia/nemotron-ocr-v2` / `v2_english/`), or pass `model_dir` to load from disk (any complete checkpoint folder; `lang` is then ignored).
 
 ```python
-from nemotron_ocr.inference.pipeline import NemotronOCR
+from nemotron_ocr.inference.pipeline_v2 import NemotronOCRV2
 
 # Default: Hugging Face v2 multilingual
-ocr = NemotronOCR()
+ocr = NemotronOCRV2()
 
-# English-optimized v2 (Hub)
-ocr_en = NemotronOCR(lang="en")
+# English v2 (Hub, word-level)
+ocr_en = NemotronOCRV2(lang="en")
 
-# Multilingual v2 explicitly (same default as NemotronOCR())
-ocr_multi = NemotronOCR(lang="multi")
+# Multilingual v2 explicitly (same default as NemotronOCRV2())
+# Uses the line-level variant.
+ocr_multi = NemotronOCRV2(lang="multi")
 
-# Local directory with detector.pth, recognizer.pth, relational.pth, charset.txt (this repo: ./checkpoints)
-ocr_local = NemotronOCR(model_dir="./checkpoints")
+# Local directory with detector.pth, recognizer.pth, relational.pth, charset.txt
+ocr_local = NemotronOCRV2(model_dir="./v2_multilingual")
 
 # Legacy v1 weights from Hub (optional)
-ocr_v1 = NemotronOCR(lang="v1")
+ocr_v1 = NemotronOCRV2(lang="v1")
 
 predictions = ocr("ocr-example-input-1.png")
 
@@ -251,7 +251,7 @@ for pred in predictions:
 **Constructor rules**
 
 - **`model_dir`**: If it contains all four checkpoint files, that directory is used and **`lang` is ignored**.
-- **`lang`** (keyword only): When weights are fetched from the Hub — `None` or `"multi"` / `"multilingual"` → [nvidia/nemotron-ocr-v2-multilingual](https://huggingface.co/nvidia/nemotron-ocr-v2-multilingual) `checkpoints/` (default); `"en"` / `"english"` → `nvidia/nemotron-ocr-v2` / `v2_english/`; `"v1"` / `"legacy"` → original v1 layout on `nvidia/nemotron-ocr-v1`.
+- **`lang`** (keyword only): When weights are fetched from the Hub — `None` or `"multi"` / `"multilingual"` → `nvidia/nemotron-ocr-v2` / `v2_multilingual/` (default); `"en"` / `"english"` → `nvidia/nemotron-ocr-v2` / `v2_english/`; `"v1"` / `"legacy"` → original v1 layout on `nvidia/nemotron-ocr-v1`.
 - If `model_dir` is set but incomplete, the client falls back to a Hub download using **`lang`** (defaulting to v2 multilingual when `lang` is `None`).
 
 ### Software Integration
@@ -270,8 +270,8 @@ for pred in predictions:
 
 ## Model Version(s)
 
-* **This repository:** Nemotron OCR **v2 multilingual** (`checkpoints/`).
-* **Related:** [nvidia/nemotron-ocr-v2](https://huggingface.co/nvidia/nemotron-ocr-v2) hosts the **v2 English** variant (`v2_english/`) and collection metadata.
+* **This repository:** Nemotron OCR v2 with both variants: `v2_english/` and `v2_multilingual/`.
+* **Hugging Face Hub:** [nvidia/nemotron-ocr-v2](https://huggingface.co/nvidia/nemotron-ocr-v2).
 
 ## **Training and Evaluation Datasets:**
 
@@ -309,23 +309,23 @@ Tables below are **reference metrics** from NVIDIA’s benchmark runs (OmniDocBe
 
 Normalized Edit Distance (NED) sample_avg on OmniDocBench (lower = better). Results follow OmniDocBench methodology (empty predictions skipped). All models evaluated in crop mode. Speed measured on a single A100 GPU.
 
-| Model | crops/s | pages/s | EN | ZH | Mixed | White | Single | Multi | Normal | Rotate90 | Rotate270 | Horizontal |
-| :--- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| PaddleOCR v5 (server) | 20.6 | 1.2 | 0.027 | 0.037 | 0.041 | 0.031 | 0.035 | 0.064 | 0.031 | 0.116 | 0.897 | 0.027 |
-| OpenOCR (server) | 17.4 | 1.5 | 0.024 | 0.033 | 0.049 | 0.027 | 0.034 | 0.061 | 0.028 | 0.042 | 0.761 | 0.034 |
-| **Nemotron OCR v2(Multilingual)** | **68.1** | **21.8** | **0.048** | **0.072** | **0.142** | **0.061** | **0.049** | **0.117** | **0.062** | **0.109** | **0.332** | **0.372** |
-| *Nemotron OCR v2 (EN)* | *74.6* | *19.9* | *0.038* | *0.830* | *0.437* | *0.348* | *0.282* | *0.572* | *0.353* | *0.232* | *0.827* | *0.893* |
-| EasyOCR | 10.3 | 0.4 | 0.095 | 0.117 | 0.326 | 0.095 | 0.179 | 0.322 | 0.110 | 0.987 | 0.979 | 0.809 |
-| Tesseract-OCR | | | 0.096 | 0.551 | 0.250 | 0.439 | 0.328 | 0.331 | 0.426 | 0.117 | 0.969 | 0.984 |
-| *Nemotron OCR v1* | *61.1* | *21.4* | *0.038* | *0.876* | *0.436* | *0.472* | *0.434* | *0.715* | *0.482* | *0.358* | *0.871* | *0.979* |
+| Model | pages/s | EN | ZH | Mixed | White | Single | Multi | Normal | Rotate90 | Rotate270 | Horizontal |
+| :--- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| PaddleOCR v5 (server) | 1.2 | 0.027 | 0.037 | 0.041 | 0.031 | 0.035 | 0.064 | 0.031 | 0.116 | 0.897 | 0.027 |
+| OpenOCR (server) | 1.5 | 0.024 | 0.033 | 0.049 | 0.027 | 0.034 | 0.061 | 0.028 | 0.042 | 0.761 | 0.034 |
+| **Nemotron OCR v2 (multilingual)** | **21.8** | **0.048** | **0.072** | **0.142** | **0.061** | **0.049** | **0.117** | **0.062** | **0.109** | **0.332** | **0.372** |
+| *Nemotron OCR v2 (EN)* | *19.9* | *0.038* | *0.830* | *0.437* | *0.348* | *0.282* | *0.572* | *0.353* | *0.232* | *0.827* | *0.893* |
+| EasyOCR | 0.4 | 0.095 | 0.117 | 0.326 | 0.095 | 0.179 | 0.322 | 0.110 | 0.987 | 0.979 | 0.809 |
+| Tesseract-OCR | | 0.096 | 0.551 | 0.250 | 0.439 | 0.328 | 0.331 | 0.426 | 0.117 | 0.969 | 0.984 |
+| *Nemotron OCR v1* | *21.4* | *0.038* | *0.876* | *0.436* | *0.472* | *0.434* | *0.715* | *0.482* | *0.358* | *0.871* | *0.979* |
 
-Column key: **crops/s** and **pages/s** are throughput using the v2 batched pipeline where measured; **EN** = English, **ZH** = Simplified Chinese, **Mixed** = English/Chinese mixed, **White/Single/Multi** = background type, **Normal/Rotate90/Rotate270/Horizontal** = text orientation.
+Column key: **pages/s** is throughput using the v2 batched pipeline where measured; **EN** = English, **ZH** = Simplified Chinese, **Mixed** = English/Chinese mixed, **White/Single/Multi** = background type, **Normal/Rotate90/Rotate270/Horizontal** = text orientation.
 
 #### [SynthDoG](https://github.com/clovaai/donut/tree/master/synthdog) Generated Benchmark Data
 
 Normalized Edit Distance (NED) page_avg on [SynthDoG](https://github.com/clovaai/donut/tree/master/synthdog) generated benchmark data (lower = better):
 
-| Language | PaddleOCR (base) | PaddleOCR (specialized) | OpenOCR (server) | Nemotron OCR v1 | *Nemotron OCR v2 (EN)* | **Nemotron OCR v2** |
+| Language | PaddleOCR (base) | PaddleOCR (specialized) | OpenOCR (server) | Nemotron OCR v1 | *Nemotron OCR v2 (EN)* | **Nemotron OCR v2 (multilingual)** |
 | :--- | ---: | ---: | ---: | ---: | ---: | ---: |
 | English | 0.117 | 0.096 | 0.105 | 0.078 | *0.079* | **0.069** |
 | Japanese | 0.201 | 0.201 | 0.586 | 0.723 | *0.765* | **0.046** |

nemotron-ocr/src/nemotron_ocr/inference/pipeline.py

@@ -41,22 +41,19 @@ DEFAULT_MERGE_LEVEL = "paragraph"
 
 # HuggingFace repositories for downloading model weights
 HF_REPO_ID = "nvidia/nemotron-ocr-v1"
-# Monorepo with per-variant folders (English under ``v2_english/``)
 HF_REPO_ID_V2 = "nvidia/nemotron-ocr-v2"
-# Multilingual weights live in this repo under ``checkpoints/`` (see Hugging Face layout)
-HF_REPO_ID_V2_MULTILINGUAL = "nvidia/nemotron-ocr-v2-multilingual"
 CHECKPOINT_FILES = ["detector.pth", "recognizer.pth", "relational.pth", "charset.txt"]
 
-# User-facing ``lang`` (see NemotronOCR ``lang``) → (repo_id, path prefix inside repo)
+# User-facing ``lang`` → (repo_id, path prefix inside repo)
 LANG_HUB_PATH: Dict[str, Tuple[str, str]] = {
     "en": (HF_REPO_ID_V2, "v2_english"),
     "english": (HF_REPO_ID_V2, "v2_english"),
-    "multi": (HF_REPO_ID_V2_MULTILINGUAL, "checkpoints"),
-    "multilingual": (HF_REPO_ID_V2_MULTILINGUAL, "checkpoints"),
+    "multi": (HF_REPO_ID_V2, "v2_multilingual"),
+    "multilingual": (HF_REPO_ID_V2, "v2_multilingual"),
     "v1": (HF_REPO_ID, "checkpoints"),
     "legacy": (HF_REPO_ID, "checkpoints"),
 }
-DEFAULT_LANG = "multi"  # v2 multilingual checkpoint from HF_REPO_ID_V2_MULTILINGUAL
+DEFAULT_LANG = "multi"
 
 
 class NemotronOCR:
@@ -65,7 +62,7 @@ class NemotronOCR:
     
     Model weights are automatically downloaded from Hugging Face Hub when no
     complete local checkpoint directory is provided. The default is Nemotron OCR
-    **v2 multilingual** (``nvidia/nemotron-ocr-v2-multilingual`` / ``checkpoints``).
+    **v2 multilingual** (``nvidia/nemotron-ocr-v2`` / ``v2_multilingual``).
 
     Automatically detects model parameters from model_config.json if available,
     otherwise falls back to defaults for backwards compatibility.
@@ -78,10 +75,11 @@ class NemotronOCR:
             being fed to the detector.  When None the value is read from
             ``model_config.json`` (key ``infer_length``), falling back to 1024.
         lang: Which checkpoint to fetch from Hugging Face when ``model_dir`` is
-            missing or incomplete: ``"en"`` / ``"english"`` (v2 English), ``"multi"`` /
-            ``"multilingual"`` (v2 multilingual, same as the default), or ``"v1"`` /
-            ``"legacy"`` (original v1 Hub layout). When ``None``, **v2 multilingual**
-            is downloaded.
+            missing or incomplete: ``"en"`` / ``"english"`` (v2 English from
+            ``nvidia/nemotron-ocr-v2`` / ``v2_english``), ``"multi"`` / ``"multilingual"``
+            (v2 multilingual from ``nvidia/nemotron-ocr-v2`` / ``v2_multilingual``, the
+            default), or ``"v1"`` / ``"legacy"`` (original v1 Hub layout).
+            When ``None``, **v2 multilingual** is downloaded.
     """
 
     def __init__(

v2_english/charset.txt

@@ -0,0 +1,857 @@
+[
+  " ",
+  "!",
+  "\"",
+  "#",
+  "$",
+  "%",
+  "&",
+  "'",
+  "(",
+  ")",
+  "*",
+  "+",
+  ",",
+  "-",
+  ".",
+  "/",
+  "0",
+  "1",
+  "2",
+  "3",
+  "4",
+  "5",
+  "6",
+  "7",
+  "8",
+  "9",
+  ":",
+  ";",
+  "<",
+  "=",
+  ">",
+  "?",
+  "@",
+  "A",
+  "B",
+  "C",
+  "D",
+  "E",
+  "F",
+  "FI",
+  "G",
+  "H",
+  "I",
+  "İ",
+  "J",
+  "K",
+  "L",
+  "M",
+  "N",
+  "O",
+  "P",
+  "Q",
+  "R",
+  "S",
+  "SS",
+  "T",
+  "U",
+  "V",
+  "W",
+  "X",
+  "Y",
+  "Z",
+  "[",
+  "\\",
+  "]",
+  "^",
+  "_",
+  "`",
+  "a",
+  "b",
+  "c",
+  "d",
+  "e",
+  "f",
+  "fi",
+  "g",
+  "h",
+  "i",
+  "i̇",
+  "j",
+  "k",
+  "l",
+  "m",
+  "n",
+  "o",
+  "p",
+  "q",
+  "r",
+  "s",
+  "ss",
+  "t",
+  "u",
+  "v",
+  "w",
+  "x",
+  "y",
+  "z",
+  "{",
+  "|",
+  "}",
+  "~",
+  "²",
+  "³",
+  "µ",
+  "¹",
+  "º",
+  "À",
+  "Á",
+  "Â",
+  "Ã",
+  "Ä",
+  "Å",
+  "Æ",
+  "Ç",
+  "È",
+  "É",
+  "Ê",
+  "Ë",
+  "Ì",
+  "Í",
+  "Î",
+  "Ï",
+  "Ð",
+  "Ñ",
+  "Ò",
+  "Ó",
+  "Ô",
+  "Õ",
+  "Ö",
+  "Ø",
+  "Ù",
+  "Ú",
+  "Û",
+  "Ü",
+  "Ý",
+  "Þ",
+  "ß",
+  "à",
+  "á",
+  "â",
+  "ã",
+  "ä",
+  "å",
+  "æ",
+  "ç",
+  "è",
+  "é",
+  "ê",
+  "ë",
+  "ì",
+  "í",
+  "î",
+  "ï",
+  "ð",
+  "ñ",
+  "ò",
+  "ó",
+  "ô",
+  "õ",
+  "ö",
+  "ø",
+  "ù",
+  "ú",
+  "û",
+  "ü",
+  "ý",
+  "þ",
+  "ÿ",
+  "Ā",
+  "ā",
+  "Ă",
+  "ă",
+  "Ą",
+  "ą",
+  "Ć",
+  "ć",
+  "Č",
+  "č",
+  "Ď",
+  "ď",
+  "Đ",
+  "đ",
+  "Ē",
+  "ē",
+  "Ė",
+  "ė",
+  "Ę",
+  "ę",
+  "Ě",
+  "ě",
+  "Ğ",
+  "ğ",
+  "Ġ",
+  "ġ",
+  "Ħ",
+  "ħ",
+  "Ĩ",
+  "ĩ",
+  "Ī",
+  "ī",
+  "İ",
+  "ı",
+  "Ķ",
+  "ķ",
+  "Ľ",
+  "ľ",
+  "Ł",
+  "ł",
+  "Ń",
+  "ń",
+  "Ņ",
+  "ņ",
+  "Ň",
+  "ň",
+  "Ŋ",
+  "ŋ",
+  "Ō",
+  "ō",
+  "Ŏ",
+  "ŏ",
+  "Ő",
+  "ő",
+  "Œ",
+  "œ",
+  "Ř",
+  "ř",
+  "Ś",
+  "ś",
+  "Ş",
+  "ş",
+  "Š",
+  "š",
+  "Ţ",
+  "ţ",
+  "Ť",
+  "ť",
+  "Ũ",
+  "ũ",
+  "Ū",
+  "ū",
+  "Ŭ",
+  "ŭ",
+  "Ů",
+  "ů",
+  "Ų",
+  "ų",
+  "Ŵ",
+  "ŵ",
+  "Ŷ",
+  "ŷ",
+  "Ÿ",
+  "Ź",
+  "ź",
+  "Ż",
+  "ż",
+  "Ž",
+  "ž",
+  "Ɓ",
+  "Ɔ",
+  "Ɖ",
+  "Ɗ",
+  "Ə",
+  "Ɛ",
+  "Ƒ",
+  "ƒ",
+  "Ɣ",
+  "Ɨ",
+  "Ɯ",
+  "Ɲ",
+  "Ɵ",
+  "Ơ",
+  "ơ",
+  "Ʀ",
+  "Ʃ",
+  "Ʈ",
+  "Ư",
+  "ư",
+  "Ʊ",
+  "Ʋ",
+  "Ʒ",
+  "ǂ",
+  "Ǎ",
+  "ǎ",
+  "Ǐ",
+  "ǐ",
+  "Ǒ",
+  "ǒ",
+  "Ǔ",
+  "ǔ",
+  "Ǫ",
+  "ǫ",
+  "Ș",
+  "ș",
+  "Ț",
+  "ț",
+  "Ʌ",
+  "ɐ",
+  "ɑ",
+  "ɒ",
+  "ɓ",
+  "ɔ",
+  "ɕ",
+  "ɖ",
+  "ɗ",
+  "ə",
+  "ɛ",
+  "ɟ",
+  "ɡ",
+  "ɢ",
+  "ɣ",
+  "ɦ",
+  "ɧ",
+  "ɨ",
+  "ɪ",
+  "ɬ",
+  "ɯ",
+  "ɲ",
+  "ɴ",
+  "ɵ",
+  "ɸ",
+  "ɻ",
+  "ɾ",
+  "ʀ",
+  "ʁ",
+  "ʂ",
+  "ʃ",
+  "ʇ",
+  "ʈ",
+  "ʊ",
+  "ʋ",
+  "ʌ",
+  "ʍ",
+  "ʎ",
+  "ʒ",
+  "ʔ",
+  "ʕ",
+  "ʘ",
+  "ʝ",
+  "ʟ",
+  "ʰ",
+  "ʲ",
+  "ʷ",
+  "ʻ",
+  "ʼ",
+  "ʾ",
+  "ʿ",
+  "ˀ",
+  "ˁ",
+  "ˈ",
+  "ˌ",
+  "ː",
+  "ˠ",
+  "ˤ",
+  "Ά",
+  "Έ",
+  "Ί",
+  "Ό",
+  "Ύ",
+  "Ώ",
+  "Α",
+  "Α͂",
+  "Β",
+  "Γ",
+  "Δ",
+  "Ε",
+  "Ζ",
+  "Η",
+  "Η͂",
+  "Θ",
+  "Ι",
+  "Ι͂",
+  "Κ",
+  "Λ",
+  "Μ",
+  "Ν",
+  "Ξ",
+  "Ο",
+  "Π",
+  "Ρ",
+  "Σ",
+  "Τ",
+  "Υ",
+  "Υ̓",
+  "Υ͂",
+  "Φ",
+  "Χ",
+  "Ψ",
+  "Ω",
+  "Ω͂",
+  "Ω͂Ι",
+  "ά",
+  "έ",
+  "ί",
+  "α",
+  "ᾶ",
+  "β",
+  "γ",
+  "δ",
+  "ε",
+  "ζ",
+  "η",
+  "ῆ",
+  "θ",
+  "ι",
+  "ῖ",
+  "κ",
+  "λ",
+  "μ",
+  "ν",
+  "ξ",
+  "ο",
+  "π",
+  "ρ",
+  "ς",
+  "σ",
+  "τ",
+  "υ",
+  "ὐ",
+  "ῦ",
+  "φ",
+  "χ",
+  "ψ",
+  "ω",
+  "ῶ",
+  "ῶι",
+  "ό",
+  "ύ",
+  "ώ",
+  "ϕ",
+  "Ё",
+  "І",
+  "Ј",
+  "А",
+  "Б",
+  "В",
+  "Г",
+  "Д",
+  "Е",
+  "Ж",
+  "З",
+  "И",
+  "Й",
+  "К",
+  "Л",
+  "М",
+  "Н",
+  "О",
+  "П",
+  "Р",
+  "С",
+  "Т",
+  "У",
+  "Х",
+  "Ц",
+  "Ч",
+  "Ш",
+  "Ъ",
+  "Ы",
+  "Ь",
+  "Э",
+  "Ю",
+  "Я",
+  "а",
+  "б",
+  "в",
+  "г",
+  "д",
+  "е",
+  "ж",
+  "з",
+  "и",
+  "й",
+  "к",
+  "л",
+  "м",
+  "н",
+  "о",
+  "п",
+  "р",
+  "с",
+  "т",
+  "у",
+  "х",
+  "ц",
+  "ч",
+  "ш",
+  "ъ",
+  "ы",
+  "ь",
+  "э",
+  "ю",
+  "я",
+  "ё",
+  "і",
+  "ј",
+  "ֵ",
+  "ֶ",
+  "ּ",
+  "א",
+  "ב",
+  "ג",
+  "ד",
+  "ו",
+  "ח",
+  "י",
+  "ל",
+  "ם",
+  "מ",
+  "נ",
+  "ס",
+  "ע",
+  "צ",
+  "ר",
+  "ש",
+  "ת",
+  "ء",
+  "أ",
+  "إ",
+  "ا",
+  "ب",
+  "ة",
+  "ت",
+  "ج",
+  "ح",
+  "خ",
+  "د",
+  "ر",
+  "ز",
+  "س",
+  "ش",
+  "ص",
+  "ط",
+  "ع",
+  "غ",
+  "ف",
+  "ق",
+  "ك",
+  "ل",
+  "م",
+  "ن",
+  "ه",
+  "و",
+  "ي",
+  "ی",
+  "ं",
+  "अ",
+  "आ",
+  "उ",
+  "क",
+  "ग",
+  "ट",
+  "ड",
+  "त",
+  "द",
+  "न",
+  "प",
+  "ब",
+  "भ",
+  "म",
+  "य",
+  "र",
+  "ल",
+  "श",
+  "ष",
+  "स",
+  "ह",
+  "ा",
+  "ि",
+  "ी",
+  "े",
+  "ो",
+  "ক",
+  "ত",
+  "ল",
+  "া",
+  "ি",
+  "க",
+  "ன",
+  "ள",
+  "ข",
+  "ง",
+  "จ",
+  "ช",
+  "ฐ",
+  "ต",
+  "ท",
+  "น",
+  "ป",
+  "พ",
+  "ร",
+  "ว",
+  "ะ",
+  "ั",
+  "า",
+  "เ",
+  "แ",
+  "ᛃ",
+  "ᛋ",
+  "ᛟ",
+  "Ḍ",
+  "ḍ",
+  "Ḥ",
+  "ḥ",
+  "Ḷ",
+  "ḷ",
+  "Ḻ",
+  "ḻ",
+  "Ṃ",
+  "ṃ",
+  "Ṅ",
+  "ṅ",
+  "Ṇ",
+  "ṇ",
+  "Ṉ",
+  "ṉ",
+  "Ṛ",
+  "ṛ",
+  "Ṟ",
+  "ṟ",
+  "Ṣ",
+  "ṣ",
+  "Ṭ",
+  "ṭ",
+  "Ṯ",
+  "ṯ",
+  "Ạ",
+  "ạ",
+  "Ả",
+  "ả",
+  "Ấ",
+  "ấ",
+  "Ầ",
+  "ầ",
+  "Ẩ",
+  "ẩ",
+  "Ẫ",
+  "ẫ",
+  "Ậ",
+  "ậ",
+  "Ắ",
+  "ắ",
+  "Ẵ",
+  "ẵ",
+  "Ặ",
+  "ặ",
+  "Ẹ",
+  "ẹ",
+  "Ế",
+  "ế",
+  "Ể",
+  "ể",
+  "Ễ",
+  "ễ",
+  "Ệ",
+  "ệ",
+  "Ị",
+  "ị",
+  "Ọ",
+  "ọ",
+  "Ỏ",
+  "ỏ",
+  "Ố",
+  "ố",
+  "Ồ",
+  "ồ",
+  "Ổ",
+  "ổ",
+  "Ỗ",
+  "ỗ",
+  "Ộ",
+  "ộ",
+  "Ớ",
+  "ớ",
+  "Ờ",
+  "ờ",
+  "Ở",
+  "ở",
+  "Ợ",
+  "ợ",
+  "Ụ",
+  "ụ",
+  "Ủ",
+  "ủ",
+  "Ứ",
+  "ứ",
+  "Ừ",
+  "ừ",
+  "Ử",
+  "ử",
+  "Ữ",
+  "ữ",
+  "Ự",
+  "ự",
+  "Ỳ",
+  "ỳ",
+  "Ỵ",
+  "ỵ",
+  "Ỹ",
+  "ỹ",
+  "ἀ",
+  "ἄ",
+  "Ἀ",
+  "Ἄ",
+  "ἐ",
+  "ἕ",
+  "Ἐ",
+  "Ἕ",
+  "ἠ",
+  "ἡ",
+  "Ἠ",
+  "Ἡ",
+  "ἰ",
+  "ἱ",
+  "Ἰ",
+  "Ἱ",
+  "ὁ",
+  "ὄ",
+  "Ὁ",
+  "Ὄ",
+  "ὐ",
+  "ὑ",
+  "Ὑ",
+  "ὡ",
+  "Ὡ",
+  "ὰ",
+  "ὲ",
+  "ὴ",
+  "ὶ",
+  "ὸ",
+  "ὺ",
+  "ὼ",
+  "ᾶ",
+  "Ὰ",
+  "ῆ",
+  "Ὲ",
+  "Ὴ",
+  "ῖ",
+  "Ὶ",
+  "ῦ",
+  "Ὺ",
+  "ῶ",
+  "ῷ",
+  "Ὸ",
+  "Ὼ",
+  "₁",
+  "₂",
+  "₃",
+  "ℓ",
+  "①",
+  "②",
+  "④",
+  "Ɑ",
+  "Ɐ",
+  "Ɒ",
+  "い",
+  "ぅ",
+  "う",
+  "お",
+  "か",
+  "き",
+  "く",
+  "ぐ",
+  "こ",
+  "し",
+  "す",
+  "せ",
+  "た",
+  "つ",
+  "ど",
+  "の",
+  "ば",
+  "ぽ",
+  "よ",
+  "ら",
+  "ん",
+  "ァ",
+  "ア",
+  "ィ",
+  "イ",
+  "ウ",
+  "ェ",
+  "エ",
+  "ォ",
+  "オ",
+  "カ",
+  "ガ",
+  "ク",
+  "グ",
+  "コ",
+  "ゴ",
+  "サ",
+  "ザ",
+  "シ",
+  "ジ",
+  "ス",
+  "ズ",
+  "セ",
+  "ゼ",
+  "ソ",
+  "タ",
+  "チ",
+  "ッ",
+  "ツ",
+  "テ",
+  "デ",
+  "ト",
+  "ド",
+  "ナ",
+  "ニ",
+  "ノ",
+  "ハ",
+  "バ",
+  "パ",
+  "ヒ",
+  "ビ",
+  "フ",
+  "ブ",
+  "プ",
+  "ベ",
+  "ペ",
+  "ボ",
+  "マ",
+  "ミ",
+  "メ",
+  "ャ",
+  "ヤ",
+  "ュ",
+  "ユ",
+  "ラ",
+  "リ",
+  "ル",
+  "レ",
+  "ロ",
+  "ワ",
+  "ン",
+  "ヴ",
+  "ー",
+  "Ɦ",
+  "Ɡ",
+  "Ɬ",
+  "Ɪ",
+  "Ʇ",
+  "Ʝ",
+  "Ʂ",
+  "거",
+  "마",
+  "막",
+  "말",
+  "사",
+  "인",
+  "전",
+  "지",
+  "짓",
+  "투",
+  "fi"
+]

v2_english/detector.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:064950b565833dfa15eaa6406a7ec9a8adc2ae159eaef9e0856f657dc0e92d2b
+size 181974624

v2_english/model_config.json

@@ -0,0 +1,18 @@
+{
+  "num_tokens": 858,
+  "max_width": 32,
+  "sequence_length": 32,
+  "scope": 2048,
+  "coordinate_mode": "RBOX",
+  "backbone": "regnet_x_8gf",
+  "charset_size": 855,
+  "recognizer_variant": "prenorm",
+  "has_pre_norm": false,
+  "has_tx_norm": true,
+  "norm_first": true,
+  "depth": 128,
+  "num_layers": 3,
+  "nhead": 8,
+  "dim_feedforward": 1024,
+  "feature_depth": 256
+}

v2_english/recognizer.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:366af771f46bfe31cfd4876e28eeab04b4ee6b11b9c9f9b6de49f7b58799f728
+size 24550133

v2_english/relational.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:763c18416f8b5285187d90e8d71f9c11b394f7a0dfd67f3e4b529a16bf583816
+size 9044661