在 DCU 上使用 OCR 推理 pdf 解析报错

[sfeng1996]

Checklist:

• 查找历史相关issue寻求解答
• 翻阅FAQ
• 翻阅PaddleX 文档
• 确认bug是否在新版本里还未修复

描述问题

在 DCU K100_ai 使用 paddlex 启动 OCR 模型，推理图片正常，推理 pdf 文件报错，通过
paddlex --pipeline /mnt/model/paddle/OCR/config-pipeline.yaml --device dcu --input /mnt/model/803_anomaly_transformer_time_serie.pdf 报错
报错信息如下：

paddlex --pipeline /mnt/model/paddle/OCR/config-pipeline.yaml --device dcu --input /mnt/model/803_anomaly_transformer_time_serie.pdf
Creating model: ('PP-LCNet_x1_0_doc_ori', '/mnt/model/paddle/OCR/PP-LCNet_x1_0_doc_ori_infer')
Skip checking if model is supported on device because the flag `PADDLE_PDX_DISABLE_DEV_MODEL_WL` has been set.
Creating model: ('UVDoc', '/mnt/model/paddle/OCR/UVDoc_infer')
Skip checking if model is supported on device because the flag `PADDLE_PDX_DISABLE_DEV_MODEL_WL` has been set.
Creating model: ('PP-LCNet_x1_0_textline_ori', '/mnt/model/paddle/OCR/PP-LCNet_x1_0_textline_ori_infer')
Skip checking if model is supported on device because the flag `PADDLE_PDX_DISABLE_DEV_MODEL_WL` has been set.
Creating model: ('PP-OCRv5_server_det', '/mnt/model/paddle/OCR/PP-OCRv5_server_det_infer')
Skip checking if model is supported on device because the flag `PADDLE_PDX_DISABLE_DEV_MODEL_WL` has been set.
Creating model: ('PP-OCRv5_server_rec', '/mnt/model/paddle/OCR/PP-OCRv5_server_rec_infer')
Skip checking if model is supported on device because the flag `PADDLE_PDX_DISABLE_DEV_MODEL_WL` has been set.
Invalid address access: 0x7efabdd64000, Error code: 1.

>>>>>>>> KERNEL VMFault !!!! <<<<<<

>>>>>>>> PID: 3854, SIGNAL: 0 !!!! <<<<<<

===========================================================================================
=========> HOSTQUEUE <0x5592179030e0>: VMFault HSA QUEUE ANALYSIS <=========
HOSTQUEUE <0x5592179030e0>: get hsa queue W/R ptr: write index: 5503, read index: 5262

HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP AQL 0x7efabc924000 PACKET WITH INDEX 5262 <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: >>>>>>>> MATCH KERNEL COMMAND <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: header: 0xb02
HOSTQUEUE <0x5592179030e0>: setup: 3
HOSTQUEUE <0x5592179030e0>: workgroup: x:1024, y:1, z:1
HOSTQUEUE <0x5592179030e0>: grid: x:307200, y:1, z:1
HOSTQUEUE <0x5592179030e0>: group_segment_size: 0
HOSTQUEUE <0x5592179030e0>: private_segment_size: 0
HOSTQUEUE <0x5592179030e0>: completion_signal: 0x0
HOSTQUEUE <0x5592179030e0>: kernel_object: 0x7ef880c51640
HOSTQUEUE <0x5592179030e0>: device id: 0

HOSTQUEUE <0x5592179030e0>: >>>>>>>> FIND MATCH KERNEL DETAILS <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: kernel name: _ZN3phi5funcs30ConcatTensorWithDifferentShapeIiLi16ENS0_20PointerAndColWrapperIfiLi8EEEEEvT1_iT_S5_Pv
HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP KERNEL ARGS: size: 120 <<<<<<<<<

00 00 00 00 80 3b 02 00 20 e8 03 00 c0 94 05 00 
60 41 07 00 00 ee 08 00 a0 9a 0a 00 40 47 0c 00 
00 11 e6 3c f8 7e 00 00 00 f0 6b 4f f8 7e 00 00 
00 55 45 3c f8 7e 00 00 00 d4 5c 36 f8 7e 00 00 
00 18 bc 35 f8 7e 00 00 00 5c 1b 35 f8 7e 00 00 
00 a0 7a 34 f8 7e 00 00 e1 dc c0 bd fa 7e 00 00 
08 00 00 00 06 00 00 00 40 47 0c 00 00 00 00 00 
00 48 91 25 f8 7e 00 00 

HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP KERNEL ARGS PTR INFO <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: ptr arg index: 4, ptr: 0x7ef83ce61100
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef83be00000, size byte: 31219968
HOSTQUEUE <0x5592179030e0>: ptr arg index: 5, ptr: 0x7ef84f6bf000
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef84f200000, size byte: 15510528
HOSTQUEUE <0x5592179030e0>: ptr arg index: 6, ptr: 0x7ef83c455500
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef83be00000, size byte: 31219968
HOSTQUEUE <0x5592179030e0>: ptr arg index: 7, ptr: 0x7ef8365cd400
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef834400000, size byte: 124518400
HOSTQUEUE <0x5592179030e0>: ptr arg index: 8, ptr: 0x7ef835bc1800
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef834400000, size byte: 124518400
HOSTQUEUE <0x5592179030e0>: ptr arg index: 9, ptr: 0x7ef8351b5c00
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef834400000, size byte: 124518400
HOSTQUEUE <0x5592179030e0>: ptr arg index: 10, ptr: 0x7ef8347aa000
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef834400000, size byte: 124518400
HOSTQUEUE <0x5592179030e0>: ptr arg index: 14, ptr: 0x7ef825914800
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef822c00000, size byte: 124518400

HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP AQL 0x7efabc924000 PACKET WITH INDEX 5263 <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: >>>>>>>> MATCH KERNEL COMMAND <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: header: 0xb02
HOSTQUEUE <0x5592179030e0>: setup: 3
HOSTQUEUE <0x5592179030e0>: workgroup: x:256, y:1, z:1
HOSTQUEUE <0x5592179030e0>: grid: x:18432, y:4, z:6
HOSTQUEUE <0x5592179030e0>: group_segment_size: 32768
HOSTQUEUE <0x5592179030e0>: private_segment_size: 0
HOSTQUEUE <0x5592179030e0>: completion_signal: 0x0
HOSTQUEUE <0x5592179030e0>: kernel_object: 0x7ef9116f5640
HOSTQUEUE <0x5592179030e0>: device id: 0

HOSTQUEUE <0x5592179030e0>: >>>>>>>> FIND MATCH KERNEL DETAILS <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: kernel name: Cijk_Ailk_Bljk_S_MT64x64x32_SE_K1_TT4_4_WG16_16_1
HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP KERNEL ARGS: size: 152 <<<<<<<<<

00 48 91 25 f8 7e 00 00 00 b0 ef ac f9 7e 00 00 
00 c1 0f 3c f8 7e 00 00 00 00 00 00 00 00 00 00 
00 1d 31 00 00 00 00 00 00 c0 02 00 00 00 00 00 
00 00 80 3f dc 11 00 00 00 dc 11 00 dc 11 00 00 
00 dc 11 00 dc 11 00 00 00 1d 31 00 c0 02 00 00 
00 00 00 00 dc 11 00 00 00 01 00 00 06 00 00 00 
c0 02 00 00 00 00 00 00 48 00 00 00 04 00 00 00 
04 00 00 00 01 00 00 00 01 00 00 80 00 00 00 00 
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
30 00 00 00 00 00 00 00 

HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP KERNEL ARGS PTR INFO <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: ptr arg index: 0, ptr: 0x7ef825914800
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef822c00000, size byte: 124518400
HOSTQUEUE <0x5592179030e0>: ptr arg index: 1, ptr: 0x7ef9acefb000
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef9acefb000, size byte: 720896
HOSTQUEUE <0x5592179030e0>: ptr arg index: 2, ptr: 0x7ef83c0fc100
HOSTQUEUE <0x5592179030e0>: origin ptr: 0x7ef83be00000, size byte: 31219968

HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP AQL 0x7efabc924000 PACKET WITH INDEX 5264 <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: >>>>>>>> MATCH KERNEL COMMAND <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: header: 0xb02
HOSTQUEUE <0x5592179030e0>: setup: 3
HOSTQUEUE <0x5592179030e0>: workgroup: x:512, y:1, z:1
HOSTQUEUE <0x5592179030e0>: grid: x:1755648, y:1, z:1
HOSTQUEUE <0x5592179030e0>: group_segment_size: 0
HOSTQUEUE <0x5592179030e0>: private_segment_size: 0
HOSTQUEUE <0x5592179030e0>: completion_signal: 0x0
HOSTQUEUE <0x5592179030e0>: kernel_object: 0x7ef9b44e4700
HOSTQUEUE <0x5592179030e0>: device id: 0

HOSTQUEUE <0x5592179030e0>: >>>>>>>> FIND MATCH KERNEL DETAILS <<<<<<<<<
HOSTQUEUE <0x5592179030e0>: kernel name: _ZN3phi5funcs25VectorizedBroadcastKernelINS0_10AddFunctorIfEEfLi2ELi1ELi4ELi1EEEvN6common5ArrayIrPKcXT1_EEENS5_IPT0_XT2_EEENS5_IbXT1_EEEjNS5_INS_3kps7details15BroadcastConfigEXT1_EEEiiiT_
HOSTQUEUE <0x5592179030e0>: >>>>>>>> DUMP KERNEL ARGS: size: 344 <<<<<<<<<

环境

使用海光社区提供的 Docker 镜像运行：
image.sourcefind.cn:5000/dcu/admin/base/paddlepaddle:3.0.0-ubuntu22.04-dtk25.04.2-py3.10
PaddlePaddle、PaddleX 版本：

import paddle, paddlex; print(f"Paddle: {paddle.__version__}, PaddleX: {paddlex.__version__}")

Paddle: 3.0.0+das, PaddleX: 3.3.5

[cuicheng01]

如果图像是可以的，PDF不行，大概率不是环境的问题，可以看看这些PDF解析成单独的图像后再推理是否有问题呢？

[TingquanGao]

The issue has no response for a long time and will be closed. You can reopen or new another issue if are still confused.

---

From Bot