Werea-DocOCR-1B: Turkish enterprise document OCR (LightOnOCR-2-1B-base SFT)

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

chat_template.jinja

@@ -0,0 +1,26 @@
+{% set image_count = namespace(value=0) %}
+{% for message in messages %}
+{% if loop.first and message['role'] != 'system' %}
+<|im_start|>system<|im_end|>
+{% endif %}
+<|im_start|>{{ message['role'] }}
+{% if message['content'] is string %}
+{{ message['content'] }}<|im_end|>
+{% else %}
+{% for content in message['content'] %}
+{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}
+{% set image_count.value = image_count.value + 1 %}
+{% if add_vision_id %}
+Picture {{ image_count.value }}:
+{% endif -%}
+<|image_pad|>
+{%- elif 'text' in content %}
+{{ content['text'] }}
+{% endif %}
+{% endfor %}
+<|im_end|>
+{% endif %}
+{% endfor %}
+{% if add_generation_prompt %}
+<|im_start|>assistant
+{% endif %}

config.json

@@ -0,0 +1,98 @@
+{
+  "architectures": [
+    "LightOnOcrForConditionalGeneration"
+  ],
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "image_token_id": 151655,
+  "model_type": "lighton_ocr",
+  "multimodal_projector_bias": false,
+  "pad_token_id": 151643,
+  "projector_hidden_act": "gelu",
+  "spatial_merge_size": 2,
+  "text_config": {
+    "architectures": [
+      "Qwen3ForCausalLM"
+    ],
+    "attention_bias": false,
+    "attention_dropout": 0,
+    "bos_token_id": null,
+    "dtype": "bfloat16",
+    "eos_token_id": null,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 1024,
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "layer_types": [
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 16384,
+    "max_window_layers": 28,
+    "model_type": "qwen3",
+    "num_attention_heads": 16,
+    "num_hidden_layers": 28,
+    "num_key_value_heads": 8,
+    "pad_token_id": null,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "rope_theta": 1000000,
+      "rope_type": "default"
+    },
+    "sliding_window": null,
+    "tie_word_embeddings": true,
+    "use_cache": true,
+    "use_qk_norm": true,
+    "use_sliding_window": false,
+    "vocab_size": 151936
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.15.0",
+  "vision_config": {
+    "attention_dropout": 0,
+    "dtype": "bfloat16",
+    "head_dim": 64,
+    "hidden_act": "silu",
+    "hidden_size": 1024,
+    "image_size": 1540,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "model_type": "pixtral",
+    "num_attention_heads": 16,
+    "num_channels": 3,
+    "num_hidden_layers": 24,
+    "patch_size": 14,
+    "rope_parameters": {
+      "rope_theta": 10000,
+      "rope_type": "default"
+    }
+  },
+  "vision_feature_layer": -1
+}

evaluation.json

@@ -0,0 +1,60 @@
+{
+  "base_model": "lightonai/LightOnOCR-2-1B-base",
+  "train_samples": 3000,
+  "epochs": 2,
+  "optimizer_steps": 375,
+  "lr": 1e-05,
+  "eval_pages": 30,
+  "base_cer": {
+    "mean": 0.6416954423539845,
+    "median": 0.6164383561643836
+  },
+  "finetuned_cer": {
+    "mean": 0.0011501285580295681,
+    "median": 0.0010395010395010396
+  },
+  "base_pred_samples": [
+    {
+      "file": "vekaletname_test_0000.jpg",
+      "cer": 0.1850828729281768,
+      "pred_head": "# T.C. GAZIANTEP 2. NOTERLİĞİ\n\n## GENEL VEKALETNAME\n\n<table>\n  <tr>\n    <td>Yevmiye No</td>\n    <td>82083</td>\n  </tr>\n  <tr>\n    <td>Tarih</td>\n    <td>04.05.2025</td>\n  </tr>\n  <tr>\n    <td>Vekalet Veren</td>\n    <td>Ali Avcı — T.C. Kimlik No: 29378410092</td>\n  </tr>\n  <tr>\n    <td>Adresi</td>\n    <td>Orhangazi Mah. Vatan Caddesi No: 146, Odunpazarı / Eskişehir</td>\n  </tr>\n  <tr>\n    <td>Vekil"
+    },
+    {
+      "file": "vekaletname_test_0001.jpg",
+      "cer": 0.26173020527859236,
+      "pred_head": "# T.C. KAYSERI 34. NOTERLİĞİ\n\n## GENEL VEKALETNAME\n\n<table>\n  <thead>\n    <tr>\n      <th>Yevmiye No</th>\n      <th>49485</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <td>Tarih</td>\n      <td>13.11.2023</td>\n    </tr>\n    <tr>\n      <td>Vekalet Veren</td>\n      <td>Aslı Ateş — T.C. Kimlik No: 55927559264</td>\n    </tr>\n    <tr>\n      <td>Adresi</td>\n      <td>Zafer Mah. Bağdat Caddesi No: 140"
+    },
+    {
+      "file": "vekaletname_test_0002.jpg",
+      "cer": 0.2773722627737226,
+      "pred_head": "# T.C. İZMİR 30. NOTERLİĞİ\n\n## GENEL VEKALETNAME\n\n<table>\n  <thead>\n    <tr>\n      <th>Yevmiye No</th>\n      <th>94329</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <td>Tarih</td>\n      <td>18.10.2026</td>\n    </tr>\n    <tr>\n      <td>Vekalet Veren</td>\n      <td>Tolga Aslan — T.C. Kimlik No: 99153535852</td>\n    </tr>\n    <tr>\n      <td>Adresi</td>\n      <td>Fatih Mah. Gazi Bulvarı No: 171/B"
+    },
+    {
+      "file": "vekaletname_test_0003.jpg",
+      "cer": 0.20121951219512196,
+      "pred_head": "# T.C. KAYSERI 23. NOTERLİĞİ\n\n## GENEL VEKALETNAME\n\n<table>\n  <tr>\n    <td>Yevmiye No</td>\n    <td>91467</td>\n  </tr>\n  <tr>\n    <td>Tarih</td>\n    <td>02.08.2026</td>\n  </tr>\n  <tr>\n    <td>Vekalet Veren</td>\n    <td>Tolga Avcı — T.C. Kimlik No: 19760557368</td>\n  </tr>\n  <tr>\n    <td>Adresi</td>\n    <td>Cumhuriyet Mah. İstiklal Caddesi No: 137/A, Kocasinan / Kayseri</td>\n  </tr>\n  <tr>\n    <td>V"
+    }
+  ],
+  "finetuned_pred_samples": [
+    {
+      "file": "vekaletname_test_0000.jpg",
+      "cer": 0.0006906077348066298,
+      "pred_head": "T.C. GAZIANTEP 2. NOTERLİĞİ\n\n# GENEL VEKALETNAME\n\n**Yevmiye No:** 82083  \n**Tarih:** 04.05.2025  \n**Vekalet Veren:** Ali Avcı — T.C. Kimlik No: 29378410092  \n**Adresi:** Orhangazi Mah. Vatan Caddesi No: 146, Odunpazarı / Eskişehir  \n**Vekil:** Ayşe Uçar — T.C. Kimlik No: 27904330760  \n**Vekil Adresi:** Gültepe Mah. İstiklal Caddesi No: 78/A, Şahinbey / Gaziantep  \n\nYukarıda açık kimliği yazılı bul"
+    },
+    {
+      "file": "vekaletname_test_0001.jpg",
+      "cer": 0.0007331378299120235,
+      "pred_head": "T.C. KAYSERI 34. NOTERLİĞİ\n\n# GENEL VEKALETNAME\n\n**Yevmiye No:** 49485  \n**Tarih:** 13.11.2023  \n**Vekalet Veren:** Aslı Ateş — T.C. Kimlik No: 55927559264  \n**Adresi:** Zafer Mah. Bağdat Caddesi No: 140, Yüreğir / Adana  \n**Vekil:** Aslı Doğan — T.C. Kimlik No: 62917968524  \n**Vekil Adresi:** Yeşilyurt Mah. Menderes Bulvarı No: 21 Kat: 6, Konyaaltı / Antalya  \n\nYukarıda açık kimliği yazılı buluna"
+    },
+    {
+      "file": "vekaletname_test_0002.jpg",
+      "cer": 0.00072992700729927,
+      "pred_head": "T.C. İZMIR 30. NOTERLİĞİ\n\n# GENEL VEKALETNAME\n\n**Yevmiye No:** 94329  \n**Tarih:** 18.10.2026  \n**Vekalet Veren:** Tolga Aslan — T.C. Kimlik No: 99153535852  \n**Adresi:** Fatih Mah. Gazi Bulvarı No: 171/B, Tepebaşı / Eskişehir  \n**Vekil:** Ceren Özdemir — T.C. Kimlik No: 74736047816  \n**Vekil Adresi:** Atatürk Mah. Şehitler Caddesi No: 179, Yıldırım / Bursa  \n\nYukarıda açık kimliği yazılı bulunan b"
+    },
+    {
+      "file": "vekaletname_test_0003.jpg",
+      "cer": 0.0007621951219512195,
+      "pred_head": "T.C. KAYSERI 23. NOTERLİĞİ\n\n# GENEL VEKALETNAME\n\n**Yevmiye No:** 91467  \n**Tarih:** 02.08.2026  \n**Vekalet Veren:** Tolga Avcı — T.C. Kimlik No: 19760557368  \n**Adresi:** Cumhuriyet Mah. İstiklal Caddesi No: 137/A, Kocasinan / Kayseri  \n**Vekil:** Onur Çetin — T.C. Kimlik No: 85438185652  \n**Vekil Adresi:** Mimar Sinan Mah. Gazi Bulvarı No: 108/B, Nilüfer / Bursa  \n\nYukarıda açık kimliği yazılı bu"
+    }
+  ]
+}

generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "temperature": 0.2,
+  "top_k": 0,
+  "top_p": 0.9,
+  "transformers_version": "5.15.0"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce5d6164c878ab2a513ecbfe4149feb3d8ea6ebef66a9003a6e6b94f1f850c3
+size 2011367496

processor_config.json

@@ -0,0 +1,30 @@
+{
+  "image_processor": {
+    "data_format": "channels_first",
+    "default_to_square": true,
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.48145466,
+      0.4578275,
+      0.40821073
+    ],
+    "image_processor_type": "PixtralImageProcessor",
+    "image_std": [
+      0.26862954,
+      0.26130258,
+      0.27577711
+    ],
+    "patch_size": 14,
+    "resample": 3,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "longest_edge": 1540
+    }
+  },
+  "patch_size": 14,
+  "processor_class": "LightOnOcrProcessor",
+  "spatial_merge_size": 2
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
+size 11422654

tokenizer_config.json

@@ -0,0 +1,28 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "image_break_token": "<|vision_pad|>",
+  "image_end_token": "<|vision_end|>",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "local_files_only": false,
+  "max_length": null,
+  "model_max_length": 131072,
+  "model_specific_special_tokens": {
+    "image_break_token": "<|vision_pad|>",
+    "image_end_token": "<|vision_end|>",
+    "image_token": "<|image_pad|>"
+  },
+  "pad_to_multiple_of": null,
+  "pad_token": "<|endoftext|>",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
+  "processor_class": "LightOnOcrProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}