레거시 모드

.gitignore

@@ -4,4 +4,18 @@ __pycache__/
 *.pyc
 *.pyo
 *.pyd
-*.log
+*.log
+gemini.md
+
+test/input/
+test/output/
+
+*.pdf
+*.jpg
+*.png
+*.jpeg
+*.tiff
+*.bmp
+*.gif
+*.svg
+*.json

api.py

@@ -1,7 +1,19 @@
 import logging
 
+from config.env_setup import setup_environment
+
+# 환경 변수 설정을 최우선으로 호출
+setup_environment()
+
+# 로깅 기본 설정
+logging.basicConfig(
+    level=logging.INFO, format="%(asctime)s [%(levelname)s] %(name)s - %(message)s"
+)
+logger = logging.getLogger("startup")
+
 from fastapi import FastAPI
 from router import deepseek_router
+from services.ocr_engine import init_engine
 
 logging.basicConfig(
     level=logging.INFO, format="%(asctime)s [%(levelname)s] %(name)s - %(message)s"
@@ -14,6 +26,17 @@ app = FastAPI(
 )
 
 
+@app.on_event("startup")
+async def startup_event():
+    """FastAPI startup event handler."""
+    logging.info("Application startup...")
+    try:
+        await init_engine()
+        logging.info("vLLM engine initialized successfully.")
+    except Exception as e:
+        logging.error(f"vLLM engine init failed: {e}", exc_info=True)
+
+
 @app.get("/health/API", include_in_schema=False)
 async def health_check():
     return {"status": "API ok"}

config/model_settings.py

@@ -22,8 +22,8 @@ MODEL_PATH = "deepseek-ai/DeepSeek-OCR"  # change to your model path
 # Omnidocbench images path: run_dpsk_ocr_eval_batch.py
 
 
-INPUT_PATH = "/workspace/input"
-OUTPUT_PATH = "/workspace/output"
+INPUT_PATH = "/workspace/test/input"
+OUTPUT_PATH = "/workspace/test/output"
 # PROMPT = f"{PROMPT_TEXT.strip()}"
 PROMPT = "<image>\n<|grounding|>Convert the document to markdown."
 # PROMPT = '<image>\nFree OCR.'

router/deepseek_router.py

@@ -1,20 +1,25 @@
 import logging
 
 from fastapi import APIRouter, File, HTTPException, UploadFile
-
 from services.ocr_engine import process_document
 
 router = APIRouter(prefix="/ocr", tags=["OCR"])
 logger = logging.getLogger(__name__)
 
 
-@router.post("", description="요청된 파일에서 Deepseek OCR을 수행하고 텍스트를 추출합니다.")
-async def perform_ocr(document: UploadFile = File(..., description="OCR을 수행할 PDF 또는 이미지 파일")):
+@router.post(
+    "", description="요청된 파일에서 Deepseek OCR을 수행하고 텍스트를 추출합니다."
+)
+async def perform_ocr(
+    document: UploadFile = File(..., description="OCR을 수행할 PDF 또는 이미지 파일"),
+):
     """
     클라이언트로부터 받은 파일을 OCR 엔진에 전달하고, 추출된 텍스트를 반환합니다.
     - **document**: `multipart/form-data` 형식으로 전송된 파일.
     """
-    logger.info(f"'{document.filename}' 파일에 대한 OCR 요청 수신 (Content-Type: {document.content_type})")
+    logger.info(
+        f"'{document.filename}' 파일에 대한 OCR 요청 수신 (Content-Type: {document.content_type})"
+    )
 
     try:
         file_content = await document.read()
@@ -36,6 +41,8 @@ async def perform_ocr(document: UploadFile = File(..., description="OCR을 수
     except Exception as e:
         # 예상치 못한 서버 내부 오류
         logger.exception(f"OCR 처리 중 예상치 못한 오류 발생: {e}")
-        raise HTTPException(status_code=500, detail=f"서버 내부 오류가 발생했습니다: {e}")
+        raise HTTPException(
+            status_code=500, detail=f"서버 내부 오류가 발생했습니다: {e}"
+        )
     finally:
         await document.close()

services/deepseek_ocr.py

@@ -9,11 +9,17 @@ import torch.nn as nn
 from addict import Dict
 
 # import time
-from config import BASE_SIZE, CROP_MODE, IMAGE_SIZE, PRINT_NUM_VIS_TOKENS, PROMPT
-from deepencoder.build_linear import MlpProjector
-from deepencoder.clip_sdpa import build_clip_l
-from deepencoder.sam_vary_sdpa import build_sam_vit_b
-from process.image_process import DeepseekOCRProcessor, count_tiles
+from config.model_settings import (
+    BASE_SIZE,
+    CROP_MODE,
+    IMAGE_SIZE,
+    PRINT_NUM_VIS_TOKENS,
+    PROMPT,
+)
+from services.deepencoder.build_linear import MlpProjector
+from services.deepencoder.clip_sdpa import build_clip_l
+from services.deepencoder.sam_vary_sdpa import build_sam_vit_b
+from services.process.image_process import DeepseekOCRProcessor, count_tiles
 from transformers import BatchFeature
 from vllm.config import VllmConfig
 from vllm.model_executor import SamplingMetadata

services/ocr_engine.py

@@ -1,37 +1,45 @@
 import asyncio
 import io
-from typing import Union
+import logging
 
 import fitz
 from config.model_settings import CROP_MODE, MODEL_PATH, PROMPT
-from fastapi import UploadFile
 from PIL import Image
-from process.image_process import DeepseekOCRProcessor
 from vllm import AsyncLLMEngine, SamplingParams
 from vllm.engine.arg_utils import AsyncEngineArgs
 from vllm.model_executor.models.registry import ModelRegistry
 
 from services.deepseek_ocr import DeepseekOCRForCausalLM
+from services.process.image_process import DeepseekOCRProcessor
+
+logger = logging.getLogger(__name__)
 
 # --------------------------------------------------------------------------
 # 1. 모델 및 프로세서 초기화
 # --------------------------------------------------------------------------
 
-# VLLM이 커스텀 모델을 인식하도록 등록
 ModelRegistry.register_model("DeepseekOCRForCausalLM", DeepseekOCRForCausalLM)
+_engine = None
+
+
+async def init_engine():
+    """vLLM 엔진을 비동기적으로 초기화합니다."""
+    global _engine
+    if _engine is not None:
+        return
+
+    engine_args = AsyncEngineArgs(
+        model=MODEL_PATH,
+        hf_overrides={"architectures": ["DeepseekOCRForCausalLM"]},
+        block_size=256,
+        max_model_len=8192,
+        enforce_eager=False,
+        trust_remote_code=True,
+        tensor_parallel_size=1,
+        gpu_memory_utilization=0.75,
+    )
+    _engine = AsyncLLMEngine.from_engine_args(engine_args)
 
-# VLLM 비동기 엔진 설정
-engine_args = AsyncEngineArgs(
-    model=MODEL_PATH,
-    hf_overrides={"architectures": ["DeepseekOCRForCausalLM"]},
-    block_size=256,
-    max_model_len=8192,
-    enforce_eager=False,
-    trust_remote_code=True,
-    tensor_parallel_size=1,
-    gpu_memory_utilization=0.75,
-)
-engine = AsyncLLMEngine.from_engine_args(engine_args)
 
 # 샘플링 파라미터 설정
 sampling_params = SamplingParams(
@@ -47,8 +55,11 @@ processor = DeepseekOCRProcessor()
 # 2. 핵심 처리 함수
 # --------------------------------------------------------------------------
 
+
 async def _process_single_image(image: Image.Image) -> str:
     """단일 PIL 이미지를 받아 OCR을 수행하고 텍스트를 반환합니다."""
+    if _engine is None:
+        raise RuntimeError("vLLM engine not initialized yet")
     if "<image>" not in PROMPT:
         raise ValueError("프롬프트에 '<image>' 토큰이 없어 OCR을 수행할 수 없습니다.")
 
@@ -60,12 +71,13 @@ async def _process_single_image(image: Image.Image) -> str:
     request_id = f"request-{asyncio.get_running_loop().time()}"
 
     final_output = ""
-    async for request_output in engine.generate(request, sampling_params, request_id):
+    async for request_output in _engine.generate(request, sampling_params, request_id):
         if request_output.outputs:
             final_output = request_output.outputs[0].text
 
     return final_output
 
+
 def _pdf_to_images(pdf_bytes: bytes, dpi=144) -> list[Image.Image]:
     """PDF 바이트를 받아 페이지별 PIL 이미지 리스트를 반환합니다."""
     images = []
@@ -83,6 +95,7 @@ def _pdf_to_images(pdf_bytes: bytes, dpi=144) -> list[Image.Image]:
     pdf_document.close()
     return images
 
+
 async def process_document(file_bytes: bytes, content_type: str, filename: str) -> dict:
     """
     업로드된 파일(이미지 또는 PDF)을 처리하여 OCR 결과를 반환합니다.
@@ -114,4 +127,4 @@ async def process_document(file_bytes: bytes, content_type: str, filename: str)
         raise ValueError(
             f"지원하지 않는 파일 형식입니다: {content_type}. "
             "이미지(JPEG, PNG 등) 또는 PDF 파일을 업로드해주세요."
-        )
+        )

services/process/image_process.py

@@ -3,13 +3,21 @@ from typing import List, Tuple
 
 import torch
 import torchvision.transforms as T
+from config.model_settings import (
+    BASE_SIZE,
+    IMAGE_SIZE,
+    MAX_CROPS,
+    MIN_CROPS,
+    PROMPT,
+    TOKENIZER,
+)
 from PIL import Image, ImageOps
-from transformers import AutoProcessor, BatchFeature, LlamaTokenizerFast
+from transformers import AutoProcessor, LlamaTokenizerFast
 from transformers.processing_utils import ProcessorMixin
-from config import IMAGE_SIZE, BASE_SIZE, CROP_MODE, MIN_CROPS, MAX_CROPS, PROMPT, TOKENIZER
+
 
 def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
-    best_ratio_diff = float('inf')
+    best_ratio_diff = float("inf")
     best_ratio = (1, 1)
     area = width * height
     for ratio in target_ratios:
@@ -25,37 +33,56 @@ def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_
     return best_ratio
 
 
-def count_tiles(orig_width, orig_height, min_num=MIN_CROPS, max_num=MAX_CROPS, image_size=640, use_thumbnail=False):
+def count_tiles(
+    orig_width,
+    orig_height,
+    min_num=MIN_CROPS,
+    max_num=MAX_CROPS,
+    image_size=640,
+    use_thumbnail=False,
+):
     aspect_ratio = orig_width / orig_height
 
     # calculate the existing image aspect ratio
     target_ratios = set(
-        (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
-        i * j <= max_num and i * j >= min_num)
+        (i, j)
+        for n in range(min_num, max_num + 1)
+        for i in range(1, n + 1)
+        for j in range(1, n + 1)
+        if i * j <= max_num and i * j >= min_num
+    )
     # print(target_ratios)
     target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
 
     # find the closest aspect ratio to the target
     target_aspect_ratio = find_closest_aspect_ratio(
-        aspect_ratio, target_ratios, orig_width, orig_height, image_size)
+        aspect_ratio, target_ratios, orig_width, orig_height, image_size
+    )
 
     return target_aspect_ratio
 
 
-def dynamic_preprocess(image, min_num=MIN_CROPS, max_num=MAX_CROPS, image_size=640, use_thumbnail=False):
+def dynamic_preprocess(
+    image, min_num=MIN_CROPS, max_num=MAX_CROPS, image_size=640, use_thumbnail=False
+):
     orig_width, orig_height = image.size
     aspect_ratio = orig_width / orig_height
 
     # calculate the existing image aspect ratio
     target_ratios = set(
-        (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
-        i * j <= max_num and i * j >= min_num)
+        (i, j)
+        for n in range(min_num, max_num + 1)
+        for i in range(1, n + 1)
+        for j in range(1, n + 1)
+        if i * j <= max_num and i * j >= min_num
+    )
     # print(target_ratios)
     target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
 
     # find the closest aspect ratio to the target
     target_aspect_ratio = find_closest_aspect_ratio(
-        aspect_ratio, target_ratios, orig_width, orig_height, image_size)
+        aspect_ratio, target_ratios, orig_width, orig_height, image_size
+    )
 
     # print(target_aspect_ratio)
     # calculate the target width and height
@@ -71,7 +98,7 @@ def dynamic_preprocess(image, min_num=MIN_CROPS, max_num=MAX_CROPS, image_size=6
             (i % (target_width // image_size)) * image_size,
             (i // (target_width // image_size)) * image_size,
             ((i % (target_width // image_size)) + 1) * image_size,
-            ((i // (target_width // image_size)) + 1) * image_size
+            ((i // (target_width // image_size)) + 1) * image_size,
         )
         # split the image
         split_img = resized_img.crop(box)
@@ -83,15 +110,13 @@ def dynamic_preprocess(image, min_num=MIN_CROPS, max_num=MAX_CROPS, image_size=6
     return processed_images, target_aspect_ratio
 
 
-
-
-
 class ImageTransform:
-
-    def __init__(self,
-                 mean: Tuple[float, float, float] = (0.5, 0.5, 0.5),
-                 std: Tuple[float, float, float] = (0.5, 0.5, 0.5),
-                 normalize: bool = True):
+    def __init__(
+        self,
+        mean: Tuple[float, float, float] = (0.5, 0.5, 0.5),
+        std: Tuple[float, float, float] = (0.5, 0.5, 0.5),
+        normalize: bool = True,
+    ):
         self.mean = mean
         self.std = std
         self.normalize = normalize
@@ -129,28 +154,28 @@ class DeepseekOCRProcessor(ProcessorMixin):
         ignore_id: int = -100,
         **kwargs,
     ):
-
         # self.candidate_resolutions = candidate_resolutions # placeholder no use
         self.image_size = IMAGE_SIZE
         self.base_size = BASE_SIZE
         # self.patch_size = patch_size
-        self.patch_size = 16 
+        self.patch_size = 16
         self.image_mean = image_mean
         self.image_std = image_std
         self.normalize = normalize
         # self.downsample_ratio = downsample_ratio
         self.downsample_ratio = 4
 
-        self.image_transform = ImageTransform(mean=image_mean, std=image_std, normalize=normalize)
-
+        self.image_transform = ImageTransform(
+            mean=image_mean, std=image_std, normalize=normalize
+        )
 
         self.tokenizer = tokenizer
         # self.tokenizer = add_special_token(tokenizer)
-        self.tokenizer.padding_side = 'left'  # must set this，padding side with make a difference in batch inference
+        self.tokenizer.padding_side = "left"  # must set this，padding side with make a difference in batch inference
 
         # add the pad_token as special token to use 'tokenizer.pad_token' and 'tokenizer.pad_token_id'
         if self.tokenizer.pad_token is None:
-            self.tokenizer.add_special_tokens({'pad_token': pad_token})
+            self.tokenizer.add_special_tokens({"pad_token": pad_token})
 
         # add image token
         # image_token_id = self.tokenizer.vocab.get(image_token)
@@ -186,9 +211,6 @@ class DeepseekOCRProcessor(ProcessorMixin):
             **kwargs,
         )
 
-
-    
-
     # def select_best_resolution(self, image_size):
     #     # used for cropping
     #     original_width, original_height = image_size
@@ -264,13 +286,21 @@ class DeepseekOCRProcessor(ProcessorMixin):
                 - num_image_tokens (List[int]): the number of image tokens
         """
 
-        assert (prompt is not None and images is not None
-                ), "prompt and images must be used at the same time."
+        assert (
+            prompt is not None and images is not None
+        ), "prompt and images must be used at the same time."
 
         sft_format = prompt
 
-        input_ids, pixel_values, images_crop, images_seq_mask, images_spatial_crop, num_image_tokens, _ = images[0]
-
+        (
+            input_ids,
+            pixel_values,
+            images_crop,
+            images_seq_mask,
+            images_spatial_crop,
+            num_image_tokens,
+            _,
+        ) = images[0]
 
         return {
             "input_ids": input_ids,
@@ -281,7 +311,6 @@ class DeepseekOCRProcessor(ProcessorMixin):
             "num_image_tokens": num_image_tokens,
         }
 
-
         # prepare = BatchFeature(
         #     data=dict(
         #         input_ids=input_ids,
@@ -341,7 +370,12 @@ class DeepseekOCRProcessor(ProcessorMixin):
         conversation = PROMPT
         assert conversation.count(self.image_token) == len(images)
         text_splits = conversation.split(self.image_token)
-        images_list, images_crop_list, images_seq_mask, images_spatial_crop = [], [], [], []
+        images_list, images_crop_list, images_seq_mask, images_spatial_crop = (
+            [],
+            [],
+            [],
+            [],
+        )
         image_shapes = []
         num_image_tokens = []
         tokenized_str = []
@@ -368,7 +402,9 @@ class DeepseekOCRProcessor(ProcessorMixin):
                     # best_width, best_height = select_best_resolution(image.size, self.candidate_resolutions)
                     # print('image ', image.size)
                     # print('open_size:', image.size)
-                    images_crop_raw, crop_ratio = dynamic_preprocess(image, image_size=IMAGE_SIZE)
+                    images_crop_raw, crop_ratio = dynamic_preprocess(
+                        image, image_size=IMAGE_SIZE
+                    )
                     # print('crop_ratio: ', crop_ratio)
                 else:
                     # best_width, best_height = self.image_size, self.image_size
@@ -383,8 +419,11 @@ class DeepseekOCRProcessor(ProcessorMixin):
                 # print('directly resize')
                 image = image.resize((self.image_size, self.image_size))
 
-            global_view = ImageOps.pad(image, (self.base_size, self.base_size),
-                                    color=tuple(int(x * 255) for x in self.image_transform.mean))
+            global_view = ImageOps.pad(
+                image,
+                (self.base_size, self.base_size),
+                color=tuple(int(x * 255) for x in self.image_transform.mean),
+            )
             images_list.append(self.image_transform(global_view))
 
             """record height / width crop num"""
@@ -392,9 +431,6 @@ class DeepseekOCRProcessor(ProcessorMixin):
             num_width_tiles, num_height_tiles = crop_ratio
             images_spatial_crop.append([num_width_tiles, num_height_tiles])
 
-
-
-
             if num_width_tiles > 1 or num_height_tiles > 1:
                 """process the local views"""
                 # local_view = ImageOps.pad(image, (best_width, best_height),
@@ -421,15 +457,22 @@ class DeepseekOCRProcessor(ProcessorMixin):
 
             # """add image tokens"""
             """add image tokens"""
-            num_queries = math.ceil((self.image_size // self.patch_size) / self.downsample_ratio)
-            num_queries_base = math.ceil((self.base_size // self.patch_size) / self.downsample_ratio)
+            num_queries = math.ceil(
+                (self.image_size // self.patch_size) / self.downsample_ratio
+            )
+            num_queries_base = math.ceil(
+                (self.base_size // self.patch_size) / self.downsample_ratio
+            )
 
-
-            tokenized_image = ([self.image_token_id] * num_queries_base + [self.image_token_id]) * num_queries_base
+            tokenized_image = (
+                [self.image_token_id] * num_queries_base + [self.image_token_id]
+            ) * num_queries_base
             tokenized_image += [self.image_token_id]
             if num_width_tiles > 1 or num_height_tiles > 1:
-                tokenized_image += ([self.image_token_id] * (num_queries * num_width_tiles) + [self.image_token_id]) * (
-                            num_queries * num_height_tiles)
+                tokenized_image += (
+                    [self.image_token_id] * (num_queries * num_width_tiles)
+                    + [self.image_token_id]
+                ) * (num_queries * num_height_tiles)
             tokenized_str += tokenized_image
             images_seq_mask += [True] * len(tokenized_image)
             num_image_tokens.append(len(tokenized_image))
@@ -447,10 +490,9 @@ class DeepseekOCRProcessor(ProcessorMixin):
             tokenized_str = tokenized_str + [self.eos_id]
             images_seq_mask = images_seq_mask + [False]
 
-        assert len(tokenized_str) == len(
-            images_seq_mask), f"tokenize_with_images func: tokenized_str's length {len(tokenized_str)} is not equal to imags_seq_mask's length {len(images_seq_mask)}"
-        
-
+        assert (
+            len(tokenized_str) == len(images_seq_mask)
+        ), f"tokenize_with_images func: tokenized_str's length {len(tokenized_str)} is not equal to imags_seq_mask's length {len(images_seq_mask)}"
 
         masked_tokenized_str = []
         for token_index in tokenized_str:
@@ -459,17 +501,21 @@ class DeepseekOCRProcessor(ProcessorMixin):
             else:
                 masked_tokenized_str.append(self.ignore_id)
 
-        assert len(tokenized_str) == len(images_seq_mask) == len(masked_tokenized_str), \
-            (f"tokenized_str's length {len(tokenized_str)}, input_ids' length {len(masked_tokenized_str)}, "
-             f"imags_seq_mask's length {len(images_seq_mask)}, are not equal")
+        assert (
+            len(tokenized_str) == len(images_seq_mask) == len(masked_tokenized_str)
+        ), (
+            f"tokenized_str's length {len(tokenized_str)}, input_ids' length {len(masked_tokenized_str)}, "
+            f"imags_seq_mask's length {len(images_seq_mask)}, are not equal"
+        )
 
         input_ids = torch.LongTensor(tokenized_str)
         target_ids = torch.LongTensor(masked_tokenized_str)
         images_seq_mask = torch.tensor(images_seq_mask, dtype=torch.bool)
 
         # set input_ids < 0 | input_ids == self.image_token_id as ignore_id
-        target_ids[(input_ids < 0) |
-                   (input_ids == self.image_token_id)] = self.ignore_id
+        target_ids[(input_ids < 0) | (input_ids == self.image_token_id)] = (
+            self.ignore_id
+        )
         input_ids[input_ids < 0] = self.pad_id
 
         inference_mode = True
@@ -484,19 +530,32 @@ class DeepseekOCRProcessor(ProcessorMixin):
         if len(images_list) == 0:
             pixel_values = torch.zeros((1, 3, self.base_size, self.base_size))
             images_spatial_crop = torch.zeros((1, 1), dtype=torch.long)
-            images_crop = torch.zeros((1, 3, self.image_size, self.image_size)).unsqueeze(0)
+            images_crop = torch.zeros(
+                (1, 3, self.image_size, self.image_size)
+            ).unsqueeze(0)
         else:
             pixel_values = torch.stack(images_list, dim=0)
             images_spatial_crop = torch.tensor(images_spatial_crop, dtype=torch.long)
             if images_crop_list:
                 images_crop = torch.stack(images_crop_list, dim=0).unsqueeze(0)
             else:
-                images_crop = torch.zeros((1, 3, self.image_size, self.image_size)).unsqueeze(0)
+                images_crop = torch.zeros(
+                    (1, 3, self.image_size, self.image_size)
+                ).unsqueeze(0)
 
         input_ids = input_ids.unsqueeze(0)
 
-        
-        return [[input_ids, pixel_values, images_crop, images_seq_mask, images_spatial_crop, num_image_tokens, image_shapes]]
+        return [
+            [
+                input_ids,
+                pixel_values,
+                images_crop,
+                images_seq_mask,
+                images_spatial_crop,
+                num_image_tokens,
+                image_shapes,
+            ]
+        ]
 
 
 AutoProcessor.register("DeepseekVLV2Processor", DeepseekOCRProcessor)

services/process/ngram_norepeat.py

@@ -1,40 +1,47 @@
+from typing import List
+
 import torch
 from transformers import LogitsProcessor
-from transformers.generation.logits_process import _calc_banned_ngram_tokens
-from typing import List, Set
 
 
 class NoRepeatNGramLogitsProcessor(LogitsProcessor):
-
-    def __init__(self, ngram_size: int, window_size: int = 100, whitelist_token_ids: set = None):
+    def __init__(
+        self, ngram_size: int, window_size: int = 100, whitelist_token_ids: set = None
+    ):
         if not isinstance(ngram_size, int) or ngram_size <= 0:
-            raise ValueError(f"`ngram_size` has to be a strictly positive integer, but is {ngram_size}")
+            raise ValueError(
+                f"`ngram_size` has to be a strictly positive integer, but is {ngram_size}"
+            )
         if not isinstance(window_size, int) or window_size <= 0:
-            raise ValueError(f"`window_size` has to be a strictly positive integer, but is {window_size}")
+            raise ValueError(
+                f"`window_size` has to be a strictly positive integer, but is {window_size}"
+            )
         self.ngram_size = ngram_size
         self.window_size = window_size
         self.whitelist_token_ids = whitelist_token_ids or set()
-    
-    def __call__(self, input_ids: List[int], scores: torch.FloatTensor) -> torch.FloatTensor:
+
+    def __call__(
+        self, input_ids: List[int], scores: torch.FloatTensor
+    ) -> torch.FloatTensor:
         if len(input_ids) < self.ngram_size:
             return scores
-        
-        current_prefix = tuple(input_ids[-(self.ngram_size - 1):])
-        
+
+        current_prefix = tuple(input_ids[-(self.ngram_size - 1) :])
+
         search_start = max(0, len(input_ids) - self.window_size)
         search_end = len(input_ids) - self.ngram_size + 1
-        
+
         banned_tokens = set()
         for i in range(search_start, search_end):
-            ngram = tuple(input_ids[i:i + self.ngram_size])
+            ngram = tuple(input_ids[i : i + self.ngram_size])
             if ngram[:-1] == current_prefix:
                 banned_tokens.add(ngram[-1])
-        
+
         banned_tokens = banned_tokens - self.whitelist_token_ids
-        
+
         if banned_tokens:
             scores = scores.clone()
             for token in banned_tokens:
                 scores[token] = -float("inf")
-        
-        return scores
+
+        return scores

test/output/result/2016-08556-211156.json

@@ -0,0 +1,12 @@
+{
+    "filename": "2016-08556-211156.pdf",
+    "model": {
+        "ocr_model": "deepseek-ocr"
+    },
+    "time": {
+        "duration_sec": "6.59",
+        "started_at": 1762395627.3137395,
+        "ended_at": 1762395633.9023185
+    },
+    "parsed": "\n수신자 한국수출입은행장 \n\n참조 EDCF Operations Department 2 \n\n제 목 방글라데시 반다주리 상수도 사업 컨설턴트 고용을 위한 문제유발 행위 불개입 확약서 \n\n1. 귀 은행의 무궁한 발전을 기원합니다. \n\n2. 표제 사업 컨설턴트 고용을 위한 제안요청서 조항에 따라 입찰 참여를 위한 \"문제유발행위 불개입 확약 서\"를 \n\n첨부와 같이 제출하오니, 참조해주시기 바랍니다. \n\n* 첨부: 문제유발행위 불개입 확약서 원본 1부. \n\n주식회사 삼안 대표이사 \n\n![](images/0_0.jpg)\n\n \n\n수신처 : Ms. Jiyoon Park, Sr. Loan Officer \n\n문서번호 201609-4495 (2016-09-22) \n\n서울 광진구 광나루로56실 85 프라임센터 34층 해외사업실 \n\n전화 02)6488-8095 \n\n담당 : 김관영 \n\nFAX 02)6488-8080 \n\n/ http://www.samaneng.com \n\n이메일 shkim5@samaneng.com<｜end▁of▁sentence｜>\n<--- Page Split --->\n"
+}

test/test.py

@@ -0,0 +1,397 @@
+import io
+import json
+import os
+import re
+import time
+
+import config.model_settings as config
+import fitz
+import img2pdf
+import numpy as np
+from config.env_setup import setup_environment
+from PIL import Image, ImageDraw, ImageFont, ImageOps
+from services.deepseek_ocr import DeepseekOCRForCausalLM
+from services.process.image_process import DeepseekOCRProcessor
+from services.process.ngram_norepeat import NoRepeatNGramLogitsProcessor
+from tqdm import tqdm
+from vllm import LLM, SamplingParams
+from vllm.model_executor.models.registry import ModelRegistry
+
+setup_environment()
+
+ModelRegistry.register_model("DeepseekOCRForCausalLM", DeepseekOCRForCausalLM)
+
+
+class Colors:
+    RED = "\033[31m"
+    GREEN = "\033[32m"
+    YELLOW = "\033[33m"
+    BLUE = "\033[34m"
+    RESET = "\033[0m"
+
+
+# --- PDF/Image Processing Functions (from run_dpsk_ocr_*.py) ---
+
+
+def pdf_to_images_high_quality(pdf_path, dpi=144):
+    images = []
+    pdf_document = fitz.open(pdf_path)
+    zoom = dpi / 72.0
+    matrix = fitz.Matrix(zoom, zoom)
+    for page_num in range(pdf_document.page_count):
+        page = pdf_document[page_num]
+        pixmap = page.get_pixmap(matrix=matrix, alpha=False)
+        Image.MAX_IMAGE_PIXELS = None
+        img_data = pixmap.tobytes("png")
+        img = Image.open(io.BytesIO(img_data))
+        if img.mode in ("RGBA", "LA"):
+            background = Image.new("RGB", img.size, (255, 255, 255))
+            background.paste(img, mask=img.split()[-1] if img.mode == "RGBA" else None)
+            img = background
+        images.append(img)
+    pdf_document.close()
+    return images
+
+
+def pil_to_pdf_img2pdf(pil_images, output_path):
+    if not pil_images:
+        return
+    image_bytes_list = []
+    for img in pil_images:
+        if img.mode != "RGB":
+            img = img.convert("RGB")
+        img_buffer = io.BytesIO()
+        img.save(img_buffer, format="JPEG", quality=95)
+        image_bytes_list.append(img_buffer.getvalue())
+    try:
+        pdf_bytes = img2pdf.convert(image_bytes_list)
+        with open(output_path, "wb") as f:
+            f.write(pdf_bytes)
+    except Exception as e:
+        print(f"Error creating PDF: {e}")
+
+
+def re_match(text):
+    pattern = r"(<\|ref\|>(.*?)<\|/ref\|><\|det\|>(.*?)<\|/det\|>)"
+    matches = re.findall(pattern, text, re.DOTALL)
+    mathes_image = [m[0] for m in matches if "<|ref|>image<|/ref|>" in m[0]]
+    mathes_other = [m[0] for m in matches if "<|ref|>image<|/ref|>" not in m[0]]
+    return matches, mathes_image, mathes_other
+
+
+def extract_coordinates_and_label(ref_text, image_width, image_height):
+    try:
+        label_type = ref_text[1]
+        cor_list = eval(ref_text[2])
+        return (label_type, cor_list)
+    except Exception as e:
+        print(f"Error extracting coordinates: {e}")
+        return None
+
+
+def draw_bounding_boxes(image, refs, jdx=None):
+    image_width, image_height = image.size
+    img_draw = image.copy()
+    draw = ImageDraw.Draw(img_draw)
+    overlay = Image.new("RGBA", img_draw.size, (0, 0, 0, 0))
+    draw2 = ImageDraw.Draw(overlay)
+    font = ImageFont.load_default()
+    img_idx = 0
+
+    for i, ref in enumerate(refs):
+        result = extract_coordinates_and_label(ref, image_width, image_height)
+        if not result:
+            continue
+
+        label_type, points_list = result
+        color = (
+            np.random.randint(0, 200),
+            np.random.randint(0, 200),
+            np.random.randint(0, 255),
+        )
+        color_a = color + (20,)
+
+        for points in points_list:
+            x1, y1, x2, y2 = [
+                int(p / 999 * (image_width if i % 2 == 0 else image_height))
+                for i, p in enumerate(points)
+            ]
+
+            if label_type == "image":
+                try:
+                    cropped = image.crop((x1, y1, x2, y2))
+                    img_filename = (
+                        f"{jdx}_{img_idx}.jpg" if jdx is not None else f"{img_idx}.jpg"
+                    )
+                    cropped.save(
+                        os.path.join(config.OUTPUT_PATH, "images", img_filename)
+                    )
+                    img_idx += 1
+                except Exception as e:
+                    print(f"Error cropping image: {e}")
+
+            width = 4 if label_type == "title" else 2
+            draw.rectangle([x1, y1, x2, y2], outline=color, width=width)
+            draw2.rectangle(
+                [x1, y1, x2, y2], fill=color_a, outline=(0, 0, 0, 0), width=1
+            )
+
+            text_x, text_y = x1, max(0, y1 - 15)
+            text_bbox = draw.textbbox((0, 0), label_type, font=font)
+            text_width, text_height = (
+                text_bbox[2] - text_bbox[0],
+                text_bbox[3] - text_bbox[1],
+            )
+            draw.rectangle(
+                [text_x, text_y, text_x + text_width, text_y + text_height],
+                fill=(255, 255, 255, 30),
+            )
+            draw.text((text_x, text_y), label_type, font=font, fill=color)
+
+    img_draw.paste(overlay, (0, 0), overlay)
+    return img_draw
+
+
+def process_image_with_refs(image, ref_texts, jdx=None):
+    return draw_bounding_boxes(image, ref_texts, jdx)
+
+
+def load_image(image_path):
+    try:
+        image = Image.open(image_path).convert("RGB")
+        return ImageOps.exif_transpose(image)
+    except Exception as e:
+        print(f"Error loading image {image_path}: {e}")
+        return None
+
+
+# --- Main OCR Processing Logic ---
+
+
+def process_pdf(llm, sampling_params, pdf_path):
+    print(f"{Colors.GREEN}Processing PDF: {pdf_path}{Colors.RESET}")
+    base_name = os.path.basename(pdf_path)
+    file_name_without_ext = os.path.splitext(base_name)[0]
+
+    images = pdf_to_images_high_quality(pdf_path)
+    if not images:
+        print(
+            f"{Colors.YELLOW}Could not extract images from {pdf_path}. Skipping.{Colors.RESET}"
+        )
+        return
+
+    batch_inputs = []
+    processor = DeepseekOCRProcessor()
+    for image in tqdm(images, desc="Pre-processing PDF pages"):
+        batch_inputs.append(
+            {
+                "prompt": config.PROMPT,
+                "multi_modal_data": {
+                    "image": processor.tokenize_with_images(
+                        images=[image], bos=True, eos=True, cropping=config.CROP_MODE
+                    )
+                },
+            }
+        )
+
+    start_time = time.time()
+    outputs_list = llm.generate(batch_inputs, sampling_params=sampling_params)
+    end_time = time.time()
+
+    contents_det = ""
+    contents = ""
+    draw_images = []
+    for i, (output, img) in enumerate(zip(outputs_list, images)):
+        content = output.outputs[0].text
+        if "<｜end of sentence｜>" in content:
+            content = content.replace("<｜end of sentence｜>", "")
+        elif config.SKIP_REPEAT:
+            continue
+
+        page_num_separator = "\n<--- Page Split --->\n"
+        contents_det += content + page_num_separator
+
+        matches_ref, matches_images, mathes_other = re_match(content)
+        result_image = process_image_with_refs(img.copy(), matches_ref, jdx=i)
+        draw_images.append(result_image)
+
+        for idx, match in enumerate(matches_images):
+            content = content.replace(match, f"![](images/{i}_{idx}.jpg)\n")
+        for match in mathes_other:
+            content = (
+                content.replace(match, "")
+                .replace("\\coloneqq", ":=")
+                .replace("\\eqqcolon", "=:")
+                .replace("\n\n\n", "\n\n")
+            )
+
+        contents += content + page_num_separator
+
+    # Save results
+    json_path = os.path.join(
+        f"{config.OUTPUT_PATH}/result", f"{file_name_without_ext}.json"
+    )
+    pdf_out_path = os.path.join(
+        config.OUTPUT_PATH, f"{file_name_without_ext}_layouts.pdf"
+    )
+
+    duration = end_time - start_time
+    output_data = {
+        "filename": base_name,
+        "model": {"ocr_model": "deepseek-ocr"},
+        "time": {
+            "duration_sec": f"{duration:.2f}",
+            "started_at": start_time,
+            "ended_at": end_time,
+        },
+        "parsed": contents,
+    }
+
+    with open(json_path, "w", encoding="utf-8") as f:
+        json.dump(output_data, f, ensure_ascii=False, indent=4)
+
+    pil_to_pdf_img2pdf(draw_images, pdf_out_path)
+    print(
+        f"{Colors.GREEN}Finished processing {pdf_path}. Results saved in {config.OUTPUT_PATH}{Colors.RESET}"
+    )
+
+
+def process_image(llm, sampling_params, image_path):
+    print(f"{Colors.GREEN}Processing Image: {image_path}{Colors.RESET}")
+    base_name = os.path.basename(image_path)
+    file_name_without_ext = os.path.splitext(base_name)[0]
+
+    image = load_image(image_path)
+    if image is None:
+        return
+
+    processor = DeepseekOCRProcessor()
+    image_features = processor.tokenize_with_images(
+        images=[image], bos=True, eos=True, cropping=config.CROP_MODE
+    )
+
+    request = {
+        "prompt": config.PROMPT,
+        "multi_modal_data": {"image": image_features},
+    }
+
+    start_time = time.time()
+    outputs = llm.generate([request], sampling_params)
+    end_time = time.time()
+    result_out = outputs[0].outputs[0].text
+
+    print(result_out)
+
+    # Save results
+    output_json_path = os.path.join(
+        f"{config.OUTPUT_PATH}/result", f"{file_name_without_ext}.json"
+    )
+    result_image_path = os.path.join(
+        config.OUTPUT_PATH, f"{file_name_without_ext}_result_with_boxes.jpg"
+    )
+
+    matches_ref, matches_images, mathes_other = re_match(result_out)
+    result_image = process_image_with_refs(image.copy(), matches_ref)
+
+    processed_text = result_out
+    for idx, match in enumerate(matches_images):
+        processed_text = processed_text.replace(match, f"![](images/{idx}.jpg)\n")
+    for match in mathes_other:
+        processed_text = (
+            processed_text.replace(match, "")
+            .replace("\\coloneqq", ":=")
+            .replace("\\eqqcolon", "=:")
+            .replace("\n\n\n", "\n\n")
+        )
+
+    duration = end_time - start_time
+    output_data = {
+        "filename": base_name,
+        "model": {"ocr_model": "deepseek-ocr"},
+        "time": {
+            "duration_sec": f"{duration:.2f}",
+            "started_at": start_time,
+            "ended_at": end_time,
+        },
+        "parsed": processed_text,
+    }
+
+    with open(output_json_path, "w", encoding="utf-8") as f:
+        json.dump(output_data, f, ensure_ascii=False, indent=4)
+
+    result_image.save(result_image_path)
+    print(
+        f"{Colors.GREEN}Finished processing {image_path}. Results saved in {config.OUTPUT_PATH}{Colors.RESET}"
+    )
+
+
+def main():
+    # --- Model Initialization ---
+    print(f"{Colors.BLUE}Initializing model...{Colors.RESET}")
+    llm = LLM(
+        model=config.MODEL_PATH,
+        hf_overrides={"architectures": ["DeepseekOCRForCausalLM"]},
+        block_size=256,
+        enforce_eager=False,
+        trust_remote_code=True,
+        max_model_len=8192,
+        swap_space=0,
+        max_num_seqs=config.MAX_CONCURRENCY,
+        tensor_parallel_size=1,
+        gpu_memory_utilization=0.9,
+        disable_mm_preprocessor_cache=True,
+    )
+
+    logits_processors = [
+        NoRepeatNGramLogitsProcessor(
+            ngram_size=20, window_size=50, whitelist_token_ids={128821, 128822}
+        )
+    ]
+
+    sampling_params = SamplingParams(
+        temperature=0.0,
+        max_tokens=8192,
+        logits_processors=logits_processors,
+        skip_special_tokens=False,
+        include_stop_str_in_output=True,
+    )
+    print(f"{Colors.BLUE}Model initialized successfully.{Colors.RESET}")
+
+    # --- File Processing ---
+    input_dir = config.INPUT_PATH
+    output_dir = config.OUTPUT_PATH
+    os.makedirs(output_dir, exist_ok=True)
+    os.makedirs(os.path.join(output_dir, "images"), exist_ok=True)
+    os.makedirs(os.path.join(output_dir, "result"), exist_ok=True)
+
+    if not os.path.isdir(input_dir):
+        print(
+            f"{Colors.RED}Error: Input directory not found at '{input_dir}'{Colors.RESET}"
+        )
+        return
+
+    print(f"Scanning for files in '{input_dir}'...")
+    for filename in sorted(os.listdir(input_dir)):
+        input_path = os.path.join(input_dir, filename)
+        if not os.path.isfile(input_path):
+            continue
+
+        file_extension = os.path.splitext(filename)[1].lower()
+
+        try:
+            if file_extension == ".pdf":
+                process_pdf(llm, sampling_params, input_path)
+            elif file_extension in [".jpg", ".jpeg", ".png", ".bmp", ".gif", ".webp"]:
+                process_image(llm, sampling_params, input_path)
+            else:
+                print(
+                    f"{Colors.YELLOW}Skipping unsupported file type: {filename}{Colors.RESET}"
+                )
+        except Exception as e:
+            print(
+                f"{Colors.RED}An error occurred while processing {filename}: {e}{Colors.RESET}"
+            )
+
+
+if __name__ == "__main__":
+    main()