DCU上的通用OCR仅支持图像文字检测，不支持PDF检测

[335079777]

描述问题

在DCU上已成功部署通用OCR产线，但是payload中只能是"image": image_data。而官方的通用OCR产线则是payload = {"file": image_data, "fileType": 1}，可以选择图像还是pdf，而dcu上的版本只能选择图像.

API_URL = "http://0.0.0.0:8080/ocr" # 服务URL
image_path = "/root/test_2.png"
output_image_path = "/root/out.jpg"
output_txt_path = "/root/result.txt"
with open(image_path, "rb") as file:
image_bytes = file.read()
image_data = base64.b64encode(image_bytes).decode("ascii")

payload = {"image": image_data} # Base64编码的文件内容或者图像URL

[Bobholamovic]

麻烦贴一下错误信息～

[335079777]

之前是在release/3.0-beta1版本，不支持。现已更新成release/3.0-rc版本

环境：jupyterlab-pytorch:2.1.0-ubuntu20.04-dtk24.04.1-py3.10-devel

Paddlepaddle版本安装如下：
python -m pip install paddlepaddle-dcu==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/dcu/
验证后应该是也可以识别DCU卡

执行下述命令报错（加白后，错误信息一致）
paddlex --serve --pipeline OCR --device gpu:0

报错信息：
Using official model (PP-LCNet_x1_0_doc_ori), the model files will be automatically downloaded and saved in /root/.paddlex/official_models.
/opt/conda/lib/python3.10/site-packages/paddle/utils/cpp_extension/extension_utils.py:711: UserWarning: No ccache found. Please be aware that recompiling all source files may be required. You can download and install ccache from: https://github.com/ccache/ccache/blob/master/doc/INSTALL.md
warnings.warn(warning_message)
Skip checking if model is supported on device because the flag PADDLE_PDX_DISABLE_DEV_MODEL_WL has been set.
Traceback (most recent call last):
File "/opt/conda/bin/paddlex", line 33, in
sys.exit(load_entry_point('paddlex', 'console_scripts', 'paddlex')())
File "/root/PaddleX/paddlex/main.py", line 26, in console_entry
main()
File "/root/PaddleX/paddlex/paddlex_cli.py", line 432, in main
serve(
File "/root/PaddleX/paddlex/paddlex_cli.py", line 324, in serve
pipeline = create_pipeline(config=pipeline_config, device=device, use_hpip=use_hpip)
File "/root/PaddleX/paddlex/inference/pipelines/init.py", line 155, in create_pipeline
pipeline = BasePipeline.get(pipeline_name)(
File "/root/PaddleX/paddlex/inference/pipelines/ocr/pipeline.py", line 64, in init
self.doc_preprocessor_pipeline = self.create_pipeline(
File "/root/PaddleX/paddlex/inference/pipelines/base.py", line 113, in create_pipeline
pipeline = create_pipeline(
File "/root/PaddleX/paddlex/inference/pipelines/init.py", line 155, in create_pipeline
pipeline = BasePipeline.get(pipeline_name)(
File "/root/PaddleX/paddlex/inference/pipelines/doc_preprocessor/pipeline.py", line 57, in init
self.doc_ori_classify_model = self.create_model(doc_ori_classify_config)
File "/root/PaddleX/paddlex/inference/pipelines/base.py", line 86, in create_model
model = create_predictor(
File "/root/PaddleX/paddlex/inference/models/init.py", line 112, in create_predictor
return BasicPredictor.get(model_name)(
File "/root/PaddleX/paddlex/inference/models/image_classification/predictor.py", line 55, in init
self.preprocessors, self.infer, self.postprocessors = self._build()
File "/root/PaddleX/paddlex/inference/models/image_classification/predictor.py", line 88, in _build
infer = StaticInfer(
File "/root/PaddleX/paddlex/inference/models/common/static_infer.py", line 125, in init
self._update_option(option)
File "/root/PaddleX/paddlex/inference/models/common/static_infer.py", line 131, in _update_option
self._reset()
File "/root/PaddleX/paddlex/inference/models/common/static_infer.py", line 148, in _reset
) = self._create()
File "/root/PaddleX/paddlex/inference/models/common/static_infer.py", line 299, in _create
predictor = create_predictor(config)
ValueError: (InvalidArgument) Fail to set gflag: conv_workspace_size_limit, please make sure the gflag exists.
[Hint: Expected success == true, but received success:0 != true:1.] (at /paddle/paddle/fluid/inference/api/analysis_predictor.cc:2287)

启动命令改为下述命令，部署成功
paddlex --serve --pipeline OCR --device dcu:0

但测试发现，实际请求中并没有用DCU卡工作，而是CPU。

[Bobholamovic]

paddlex --serve --pipeline OCR --device gpu:0这个指令试图使用GPU部署服务，如果不存在GPU设备，理论上应该不会执行成功；由于paddle底层对于DCU和GPU复用了一套处理，所以这里给出的报错信息可能不够清晰。

如果执行paddlex --serve --pipeline OCR --device dcu的话，是否可以呢？

[335079777]

paddlex --serve --pipeline OCR --device dcu这样是可以部署运行的，但是实际测试中发现，DCU的利用率为0，看起来是用CPU来跑的，没有用DCU呢。

[Bobholamovic]

这个看上去大概率不是服务化部署的问题，建议试试本地推理（API或者CLI）是否也有类似的现象，如果也存在的话，我会联系更熟悉这块的同事帮忙看看