Fix params and Update README.

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

@@ -9,7 +9,7 @@ os.environ["HIP_VISIBLE_DEVICES"] = '0'
 
 parse = argparse.ArgumentParser()
 parse.add_argument('--model_name_or_path', type=str, default='deepseek-ai/DeepSeek-OCR')
-parse.add_argument('--image_file', type=str, default='./doc/test.jpg')
+parse.add_argument('--image_file', type=str, default='./doc/test.png')
 parse.add_argument('--output_path', type=str, default='./output/')
 args = parse.parse_args()
 

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

@@ -21,8 +21,8 @@ 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 = ''
+INPUT_PATH = './doc/test.png'
+OUTPUT_PATH = './output'
 
 PROMPT = '<image>\n<|grounding|>Convert the document to markdown.'
 # PROMPT = '<image>\nFree OCR.'

DeepSeek-OCR-master/DeepSeek-OCR-vllm/run_dpsk_ocr_image.py → DeepSeek-OCR-vllm/run_dpsk_ocr_image.py

 import asyncio
 import re
 import os
+import argparse
 
-import torch
-if torch.version.cuda == '11.8':
-    os.environ["TRITON_PTXAS_PATH"] = "/usr/local/cuda-11.8/bin/ptxas"
-
-os.environ['VLLM_USE_V1'] = '0'
-os.environ["CUDA_VISIBLE_DEVICES"] = '0'
+os.environ["HIP_VISIBLE_DEVICES"] = '0'
 
 from vllm import AsyncLLMEngine, SamplingParams
 from vllm.engine.arg_utils import AsyncEngineArgs
@@ -21,8 +17,6 @@ from process.ngram_norepeat import NoRepeatNGramLogitsProcessor
 from process.image_process import DeepseekOCRProcessor
 from config import MODEL_PATH, INPUT_PATH, OUTPUT_PATH, PROMPT, CROP_MODE
 
-
-
 ModelRegistry.register_model("DeepseekOCRForCausalLM", DeepseekOCRForCausalLM)
 
 def load_image(image_path):
@@ -136,21 +130,17 @@ def draw_bounding_boxes(image, refs):
     img_draw.paste(overlay, (0, 0), overlay)
     return img_draw
 
-
 def process_image_with_refs(image, ref_texts):
     result_image = draw_bounding_boxes(image, ref_texts)
     return result_image
 
-
-
-
 async def stream_generate(image=None, prompt=''):
 
 
     engine_args = AsyncEngineArgs(
         model=MODEL_PATH,
         hf_overrides={"architectures": ["DeepseekOCRForCausalLM"]},
-        block_size=256,
+        block_size=64,
         max_model_len=8192,
         enforce_eager=False,
         trust_remote_code=True,

DeepSeek-OCR-master/DeepSeek-OCR-vllm/run_dpsk_ocr_pdf.py → DeepSeek-OCR-vllm/run_dpsk_ocr_pdf.py

@@ -3,17 +3,12 @@ import fitz
 import img2pdf
 import io
 import re
+
 from tqdm import tqdm
-import torch
 from concurrent.futures import ThreadPoolExecutor
 
-
-if torch.version.cuda == '11.8':
-    os.environ["TRITON_PTXAS_PATH"] = "/usr/local/cuda-11.8/bin/ptxas"
-os.environ['VLLM_USE_V1'] = '0'
 os.environ["CUDA_VISIBLE_DEVICES"] = '0'
 
-
 from config import MODEL_PATH, INPUT_PATH, OUTPUT_PATH, PROMPT, SKIP_REPEAT, MAX_CONCURRENCY, NUM_WORKERS, CROP_MODE
 
 from PIL import Image, ImageDraw, ImageFont
@@ -32,7 +27,7 @@ ModelRegistry.register_model("DeepseekOCRForCausalLM", DeepseekOCRForCausalLM)
 llm = LLM(
     model=MODEL_PATH,
     hf_overrides={"architectures": ["DeepseekOCRForCausalLM"]},
-    block_size=256,
+    block_size=64,
     enforce_eager=False,
     trust_remote_code=True,
     max_model_len=8192,

README.md

@@ -16,10 +16,6 @@ DeepSeek-OCR 的能力范围包括：
 - 多语言处理（中英文混合识别）
 - 物体定位（grounding 功能支持）
 
-<div align=center>
-    <img src="./doc/xxx.png"/>
-</div>
-
 ## 环境配置
 ### 硬件需求
 DCU型号：K100AI，节点数量：1台，卡数：1张。
@@ -28,7 +24,7 @@ DCU型号：K100AI，节点数量：1台，卡数：1张。
 
 ### Docker（方法一）
 ```bash
-docker pull image.sourcefind.cn:5000/dcu/admin/base/vllm:0.9.2-ubuntu22.04-dtk25.04.1-rc5-rocblas104381-0915-das1.6-py3.10-20250916-rc2-ds3.2
+docker pull image.sourcefind.cn:5000/dcu/admin/base/vllm:0.8.5-ubuntu22.04-dtk25.04.1-rc5-das1.6-py3.10-20250724
 
 docker run -it --shm-size 200g --network=host --name {docker_name} --privileged --device=/dev/kfd --device=/dev/dri --device=/dev/mkfd --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -u root -v /path/your_code_data/:/path/your_code_data/ -v /opt/hyhal/:/opt/hyhal/:ro {imageID} bash
 
@@ -52,7 +48,7 @@ DTK: 25.04.1
 python: 3.10.12
 torch: 2.5.1+das.opt1.dtk25041
 transformers: 4.46.3
-vllm: 0.9.2
+vllm: 0.8.5
 ```
 `Tips：以上dtk驱动、pytorch等DCU相关工具版本需要严格一一对应`, 其它非深度学习库参照requirements.txt安装：
 ```bash
@@ -67,25 +63,25 @@ pip install -r requirements.txt
 
 ## 推理
 ### transformers
-模型地址，测试图片路径，输出路径根据实际情况修改。
+> 模型地址，测试图片路径，输出路径根据实际情况修改。
 ```bash
-cd DeepSeek-OCR-master/DeepSeek-OCR-hf
-python run_dpsk_ocr.py --model_name_or_path=deepseek-ai/DeepSeek-OCR --image_path=./doc/test.jpg --output_path=./output
+python DeepSeek-OCR-hf/run_dpsk_ocr.py --model_name_or_path=deepseek-ai/DeepSeek-OCR --image_path=./doc/test.png --output_path=./output
 ```
 
 ### vllm
+> 模型地址，测试图片路径，输出路径请根据实际情况在`DeepSeek-OCR-vllm/config.py`中修改。
 ```bash
-cd DeepSeek-OCR-master/DeepSeek-OCR-vllm
-# image: streaming output
-python run_dpsk_ocr_image.py
+export VLLM_USE_V1=0
+# image：流式输出
+python DeepSeek-OCR-vllm/run_dpsk_ocr_image.py
 # pdf
-python run_dpsk_ocr_pdf.py
+python DeepSeek-OCR-vllm/run_dpsk_ocr_pdf.py
 ```
 
 ## result
 
 <div align=center>
-    <img src="./doc/xxx.png"/>
+    <img src="./doc/result_with_boxes_vllm.jpg"/>
 </div>
 
 ### 精度

docker/Dockerfile

-FROM image.sourcefind.cn:5000/dcu/admin/base/vllm:0.9.2-ubuntu22.04-dtk25.04.1-rc5-rocblas104381-0915-das1.6-py3.10-20250916-rc2-ds3.2
+FROM image.sourcefind.cn:5000/dcu/admin/base/vllm:0.8.5-ubuntu22.04-dtk25.04.1-rc5-das1.6-py3.10-20250724
\ No newline at end of file

model.properties

 # 模型唯一标识
-modelCode=205
+modelCode=1781
 # 模型名称
-modelName=PaddleOCR_paddle_onnxruntime
+modelName=deepseek-ocr_pytorch
 # 模型描述
-modelDescription=paddleocr_paddle_onnxruntime是一个实现字符检测和识别的模型。
+modelDescription=DeepSeek 推出了全新的视觉文本压缩模型 DeepSeek-OCR。
 # 应用场景
-appScenario=推理,训练,OCR,制造,金融,交通,教育,医疗
+appScenario=推理,OCR,制造,金融,交通,教育,医疗
 # 框架类型
-frameType=paddle,onnxruntime
+frameType=pytorch,vllm
 # 加速卡类型
 accelerateType=K100AI
\ No newline at end of file