first commit

DeepSeek-OCR-master/DeepSeek-OCR-hf/run_dpsk_ocr.py

@@ -17,8 +17,8 @@ model = model.eval().cuda().to(torch.bfloat16)
 
 # prompt = "<image>\nFree OCR. "
 prompt = "<image>\n<|grounding|>Convert the document to markdown. "
-image_file = 'your_image.jpg'
-output_path = 'your/output/dir'
+image_file = '/workspace/2025-27484-M21472.pdf'
+output_path = '/workspace/output_hf'
 
 
 

DeepSeek-OCR-master/DeepSeek-OCR-vllm/config.py

@@ -21,8 +21,10 @@ MODEL_PATH = 'deepseek-ai/DeepSeek-OCR' # change to your model path
 # .jpg, .png, .jpeg: run_dpsk_ocr_image.py; 
 # Omnidocbench images path: run_dpsk_ocr_eval_batch.py
 
-INPUT_PATH = '' 
-OUTPUT_PATH = ''
+FILE_NAME='2025-27484-M21472.pdf'
+
+INPUT_PATH = f'/workspace/2025-27484-M21472.pdf' 
+OUTPUT_PATH = '/workspace/output/'
 
 PROMPT = '<image>\n<|grounding|>Convert the document to markdown.'
 # PROMPT = '<image>\nFree OCR.'

DeepSeek-OCR-master/DeepSeek-OCR-vllm/requirements.txt

@@ -0,0 +1,10 @@
+transformers
+tokenizers
+PyMuPDF
+img2pdf
+einops
+easydict
+addict 
+Pillow
+numpy
+matplotlib

Dockerfile.hf

@@ -0,0 +1,48 @@
+# PyTorch 2.6.0 + CUDA 12.6 + cuDNN9
+FROM pytorch/pytorch:2.6.0-cuda12.6-cudnn9-devel
+
+ENV DEBIAN_FRONTEND=noninteractive \
+    HF_HOME=/workspace/.cache/huggingface \
+    CUDA_HOME=/usr/local/cuda \
+    LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH} \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PYTHONUNBUFFERED=1 \
+    HF_HUB_DISABLE_TELEMETRY=1
+
+ARG TORCH_CUDA_ARCH_LIST=80
+ENV TORCH_CUDA_ARCH_LIST=${TORCH_CUDA_ARCH_LIST}
+ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
+
+WORKDIR /workspace
+
+# 빌드 도구
+RUN apt-get update && apt-get install -y --no-install-recommends \
+      git build-essential ninja-build cmake \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN python -m pip install -U pip setuptools wheel packaging ninja
+
+# 명시 재설치(일관성)
+RUN pip install --no-cache-dir \
+      torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0
+
+# DeepSeek-OCR(HF) 호환 스택
+RUN pip install --no-cache-dir \
+      "transformers==4.43.3" \
+      "accelerate==0.33.0" \
+      "tokenizers==0.19.1" \
+      "numpy==1.26.4" \
+      "safetensors>=0.4.2" \
+      "einops" "timm>=0.9"
+
+# flash-attn 2.7.3 (CUDA 12.6에서 빌드)
+RUN pip install --no-cache-dir --no-build-isolation --no-binary=flash-attn flash-attn==2.7.3
+
+# vLLM 제거(혹시 들어오더라도 충돌 방지)
+RUN pip uninstall -y vllm || true
+
+# OpenCV ↔ numpy 1.26 호환 버전 고정
+RUN pip install --no-cache-dir "opencv-python-headless==4.8.1.78"
+
+# 앱 소스
+COPY DeepSeek-OCR-master/DeepSeek-OCR-hf/ /workspace/DeepSeek-OCR-hf/

Dockerfile.vllm

@@ -0,0 +1,41 @@
+# PyTorch 2.6.0 + CUDA 12.6 + cuDNN9
+FROM pytorch/pytorch:2.6.0-cuda12.6-cudnn9-devel
+
+# 기본 환경 변수
+ENV DEBIAN_FRONTEND=noninteractive \
+    HF_HOME=/workspace/.cache/huggingface \
+    CUDA_HOME=/usr/local/cuda \
+    LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH} \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PYTHONUNBUFFERED=1
+
+WORKDIR /workspace
+
+# 필수 빌드 도구
+RUN apt-get update && apt-get install -y --no-install-recommends \
+      git build-essential ninja-build \
+    && rm -rf /var/lib/apt/lists/*
+
+# pip 툴업
+RUN python -m pip install -U pip setuptools wheel packaging ninja
+
+# (베이스 이미지에 torch 2.6.0이 이미 포함)
+# torchvision/torchaudio 버전을 보장하려면 아래 라인을 사용하세요.
+# cu126 인덱스가 없을 수 있으므로 실패해도 빌드가 계속되게 처리합니다.
+RUN pip install --no-cache-dir \
+      torchvision==0.21.0 torchaudio==2.6.0 \
+      --extra-index-url https://download.pytorch.org/whl/cu126 || \
+    echo "Skip reinstalling torchvision/torchaudio (fallback to base image)"
+
+# 프로젝트 의존성 설치
+COPY requirements.txt /tmp/requirements.txt
+RUN pip install --no-cache-dir -r /tmp/requirements.txt
+
+# vLLM 0.8.5 고정 (CUDA 12.x용 wheel 자동 선택)
+RUN pip install --no-cache-dir vllm==0.8.5
+
+# FlashAttention (CUDA 12.6에서 빌드)
+RUN pip install --no-cache-dir flash-attn==2.7.3 --no-build-isolation
+
+# 작업 디렉터리 유지
+WORKDIR /workspace

docker-compose.yml

@@ -0,0 +1,42 @@
+services:
+  deepseek_ocr_vllm:
+    build:
+      context: .
+      dockerfile: Dockerfile.vllm
+    image: deepseek-ocr-vllm:cu126
+    container_name: deepseek_ocr_vllm
+    working_dir: /workspace
+    volumes:
+      - ./DeepSeek-OCR-master/DeepSeek-OCR-vllm:/workspace
+    gpus: all
+    shm_size: "8g"
+    ipc: "host"
+    environment:
+      - HF_HOME=/workspace/.cache/huggingface
+      - CUDA_HOME=/usr/local/cuda
+      - LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
+      - PIP_DISABLE_PIP_VERSION_CHECK=1
+      - PYTHONUNBUFFERED=1
+    tty: true
+    entrypoint: ["/bin/bash"]
+
+  # deepseek_ocr_hf:
+  #   build:
+  #     context: .
+  #     dockerfile: Dockerfile.hf
+  #   image: deepseek-ocr-hf:cu126
+  #   container_name: deepseek_ocr_hf
+  #   working_dir: /workspace
+  #   volumes:
+  #     - ./DeepSeek-OCR-hf:/workspace
+  #   gpus: all
+  #   shm_size: "8g"
+  #   ipc: "host"
+  #   environment:
+  #     - HF_HOME=/workspace/.cache/huggingface
+  #     - CUDA_HOME=/usr/local/cuda
+  #     - LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
+  #     - PIP_DISABLE_PIP_VERSION_CHECK=1
+  #     - PYTHONUNBUFFERED=1
+  #   tty: true
+  #   entrypoint: ["/bin/bash"]

requirements.txt

@@ -1,5 +1,5 @@
-transformers==4.46.3
-tokenizers==0.20.3
+transformers
+tokenizers
 PyMuPDF
 img2pdf
 einops
@@ -7,3 +7,4 @@ easydict
 addict 
 Pillow
 numpy
+matplotlib