main/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell

[giscus[bot]]

main/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)

http://www.paddleocr.ai/main/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.html

[HK-SOURCE-beep]

你好，请问PaddleOCR-vl生成的json文件，有自带的解析成word方法吗

[lilong1988]

之前的版本通过PaddleX 高稳定性服务化部署方案基于 NVIDIA Triton Inference Server 打造，可以修改config_gpu.pbtxt指定每一设备上放置的实例数量。
通过Docker Compose 部署 PaddleOCR-VL-1.5，如何将PaddleOCR服务（paddleocr-vl-api）在一张显卡上起多个实例？

[HK-SOURCE-beep]

用了下感觉1.5b在还不完善，手写文字检测不如0.9b

[ztommy281-del]

你好，使用纯cpu跑，解析一个png大概需要多久，我本地解析很慢，大概需要几分钟，如果使用GPU会很快吗

[huotong1212]

我在开启版面分析检测后，如果印章区域的内容与table部分也重叠，则table的部分未识别，如果解决这个问题？

[lhy531]

文档里提到的参数markdown_ignore_labels，在框架中并没有得到支持ValueError: Unknown argument: markdown_ignore_labels

[lhy531]

ModuleNotFoundError: No module named 'paddleformers.trl'

[donga-q]

"您可以分别从 此处 和 此处 获取 Compose 文件与环境变量配置文件并下载到本地，然后在刚刚下载的文件所在目录下执行以下命令启动服务器"
这两个此处的连接点进去为空

[cty-ai]

1.6版本高性能服务化部署报错：
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Max concurrent inference requests: 16
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Inference timeout: 600s
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Inference timeout: 600s
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Max concurrent non-inference requests: 64
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Triton URL: paddleocr-vl-tritonserver:8001
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Inference timeout: 600s
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Max concurrent inference requests: 16
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Max concurrent non-inference requests: 64
paddleocr-vl-api | 2026-06-02 14:00:56,539 - app - _lifespan - INFO - Inference timeout: 600s
paddleocr-vl-api | 2026-06-02 14:00:56,541 - app - _lifespan - INFO - Gateway initialized successfully
paddleocr-vl-api | 2026-06-02 14:00:56,541 - app - _lifespan - INFO - Gateway initialized successfully
paddleocr-vl-api | INFO: Application startup complete.
paddleocr-vl-api | INFO: Application startup complete.
paddleocr-vl-api | 2026-06-02 14:00:56,541 - app - _lifespan - INFO - Gateway initialized successfully
paddleocr-vl-api | 2026-06-02 14:00:56,541 - app - _lifespan - INFO - Gateway initialized successfully
paddleocr-vl-api | INFO: Application startup complete.
paddleocr-vl-api | INFO: Application startup complete.
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:59278 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:47776 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:53472 - "GET /health HTTP/1.1" 200 OK
paddleocr-vl-api | 2026-06-02 14:01:30,757 - app - _process_triton_request - INFO - Processing 'layout-parsing' request ecf7e5d3-5509-479a-ad9e-2b37aceca52a
paddleocr-vl-tritonserver | [ INFO] [2026-06-02 14:01:30,765] [ec1acde8d021413395d70929e8fad7cf] [5abc704ee7c0495a963178ed1f2d188c] - Received batch of size 1
paddleocr-vl-tritonserver | [ INFO] [2026-06-02 14:01:30,807] [ec1acde8d021413395d70929e8fad7cf] [5abc704ee7c0495a963178ed1f2d188c] - Request ecf7e5d3-5509-479a-ad9e-2b37aceca52a received
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:57250 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) ERROR 06-02 14:01:32 [serving_chat.py:176] Error with model error=ErrorInfo(message='The model PaddleOCR-VL-1.5-0.9B does not exist.', type='NotFoundError', param=None, code=404)
paddleocr-vlm-server | (APIServer pid=1) ERROR 06-02 14:01:32 [serving_chat.py:176] Error with model error=ErrorInfo(message='The model PaddleOCR-VL-1.5-0.9B does not exist.', type='NotFoundError', param=None, code=404)
paddleocr-vlm-server | (APIServer pid=1) INFO: 172.18.0.3:37920 - "POST /v1/chat/completions HTTP/1.1" 404 Not Found
paddleocr-vlm-server | (APIServer pid=1) INFO: 172.18.0.3:37922 - "POST /v1/chat/completions HTTP/1.1" 404 Not Found
paddleocr-vl-tritonserver | [ ERROR] [2026-06-02 14:01:32,717] [ec1acde8d021413395d70929e8fad7cf] [5abc704ee7c0495a963178ed1f2d188c] - Unhandled exception
paddleocr-vl-tritonserver | Traceback (most recent call last):
paddleocr-vl-tritonserver | File "/paddlex/py310/lib/python3.10/site-packages/paddlex_hps_server/base_model.py", line 122, in execute
paddleocr-vl-tritonserver | result_or_output_lst = self.run_batch(
paddleocr-vl-tritonserver | File "/paddlex/var/paddlex_model_repo/layout-parsing/1/model.py", line 126, in run_batch
paddleocr-vl-tritonserver | preds = list(
paddleocr-vl-tritonserver | File "/paddlex/py310/lib/python3.10/site-packages/paddlex/inference/pipelines/_parallel.py", line 139, in predict
paddleocr-vl-tritonserver | yield from self._pipeline.predict(

paddleocr-vl-tritonserver | File "/paddlex/py310/lib/python3.10/site-packages/paddlex/inference/pipelines/paddleocr_vl/pipeline.py", line 889, in predict
paddleocr-vl-api | 2026-06-02 14:01:32,726 - app - _process_triton_request - WARNING - Triton returned error for 'layout-parsing' request ecf7e5d3-5509-479a-ad9e-2b37aceca52a: Internal server error

paddleocr-vl-api | INFO: 172.18.0.1:59222 - "POST /layout-parsing HTTP/1.1" 500 Internal Server Error
paddleocr-vl-tritonserver | raise RuntimeError(
paddleocr-vl-tritonserver | RuntimeError: Exception from the 'vlm' worker: Error code: 404 - {'error': {'message': 'The model PaddleOCR-VL-1.5-0.9B does not exist.', 'type': 'NotFoundError', 'param': None, 'code': 404}}
paddleocr-vl-tritonserver | [ INFO] [2026-06-02 14:01:32,724] [ec1acde8d021413395d70929e8fad7cf] [5abc704ee7c0495a963178ed1f2d188c] - Time taken: 1958.110 ms
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:33886 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:40290 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:51452 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:39482 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:44388 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:39916 - "GET /health HTTP/1.1" 200 OK
paddleocr-vlm-server | (APIServer pid=1) INFO: 127.0.0.1:50070 - "GET /health HTTP/1.1" 200 OK

[1766868838]

你好，请问PaddleOCR-vl本地模型会将星号包裹成 $ ^{{**}} $，有参数能配置吗