昇腾社区首页
支持
我的论坛主页
论坛首页
/
MindSpore-lite在云侧量化卡住失败
收藏回复举报
MindSpore-lite在云侧量化卡住失败
发表于2024-12-25 08:31:48
0 查看

1 系统环境

硬件环境(Ascend/GPU/CPU): Ascend/CPU

MindSpore版本: mindspore=2.2.14

执行模式（PyNative/ Graph）:不限

Python版本: Python=3.9

操作系统平台: linux

2 报错信息

2.1 问题描述

驱动版本npu-smi version: 23.0.1，mindspore-lite版本2.2.14；模型为yolox-s进行训练后量化，采用全量化方法，会出现卡死情况。相同模型和条件，如果将设备从Ascend修改为CPU进行量化则能够成功，输出结果也是正确的，但是推理速度会变慢几十倍。

2.2 转换命令：

./converter_lite --fmk=ONNX --saveType=MINDIR --optimize=ascend_oriented --modelFile=./yolox-s.onnx --outputFile=./yolox --inputShape="images:1,3,640,640" --inputDataFormat=NCHW --configFile=./config.cfg

2.3 脚本信息

[common_quant_param]

quant_type=FULL_QUANT

bit_num=8



[data_preprocess_param]

calibrate_path=data:./images/

calibrate_size=100

input_type=IMAGE

image_to_format=BGR

normalize_mean=[0, 0, 0]

normalize_std=[1, 1, 1]

resize_width=640

resize_height=640

resize_method=LINEAR

center_crop_width=640

center_crop_height=640 



[full_quant_param]

activation_quant_method=MAX_MIN

target_device=ASCEND

2.4 报错信息

[WARNING] LITE(1467645,ffffaf5b1020,converter_lite):2024-11-26-08:46:10.478.008 [mindspore/lite/tools/converter/quantizer/quant_strategy.cc:179]

CanOpFullQuantized] node:Sigmoid_265_post type:Transpose will not quantify.                                                                      

[WARNING] LITE(1467645,ffffaf5b1020,converter_lite):2024-11-26-08:46:10.478.025 [mindspore/lite/tools/converter/quantizer/quant_strategy.cc:179]

CanOpFullQuantized] node:Concat_266 type:Concat will not quantify.                                                                               

[WARNING] LITE(1467645,ffffaf5b1020,converter_lite):2024-11-26-08:46:10.613.436 [mindspore/lite/src/extendrt/numa_adapter.cc:36] NUMAAdapter] Ope

n libnuma.so failed!try libnuma.so.1 again.                                                                                                      

[ERROR] LITE(1467645,ffffaf5b1020,converter_lite):2024-11-26-08:46:10.623.237 [mindspore/lite/src/extendrt/mindir_loader/mindir_model/mindir_mode

l_loader.cc:39] ImportModel] "MindirModelLoader: Import model failed, please check the correctness of the file."                                

[ERROR] LITE(1467645,ffffad199900,converter_lite):2024-11-26-08:46:59.766.738 [mindspore/lite/src/extendrt/mindir_loader/mindir_model/mindir_mode

l_loader.cc:39] ImportModel] "MindirModelLoader: Import model failed, please check the correctness of the file."

开启日志后信息如下：

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.875.523 [mindspore/lite/src/litert/lite_session.cc:934] InitSharedThreadPool ] runner id:   enable_shared_pool: 0  workers_num: -1  thread_num_limit: -1  remaining_thread_num: -1

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.923.584 [mindspore/lite/src/litert/lite_session.cc:982] AscendInit] There is  no Ascend device type.

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.923.694 [mindspore/lite/src/litert/pack_weight_manager.cc:89] GenModelID] ge nerate model id: model_2

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.923.731 [mindspore/lite/src/litert/pack_weight_manager.cc:120] InitPackWeigh tManager] model use share pack weight.

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.926.350 [mindspore/lite/src/litert/pack_weight_manager.cc:137] GetSharedMode lBuf] model use share pack weight.

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.926.904 [mindspore/lite/src/litert/lite_model.cc:598] ImportFromBuffer] impo rt model from model loader  [ERROR] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.926.954 [mindspore/lite/src/extendrt/mindir_loader/mindir_model/mindir_mode l_loader.cc:39] ImportModel] "MindirModelLoader: Import model failed, please check the correctness of the file."

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.926.979 [mindspore/lite/src/litert/lite_model.cc:605] ImportFromBuffer] impo rt model from lite model

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.927.709 [mindspore/lite/src/litert/lite_model.cc:430] GenerateModelByVersion ] MindSpore Lite inference version: MindSpore Lite 2.2.14

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.929.944 [mindspore/lite/src/litert/pass/online_fusion/online_fusion_pass_reg istry.cc:35] DoOnlineFusionPass] do online fusion pass : DoCastGatherReduceFusionPass

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.930.088 [mindspore/lite/src/litert/pass/online_fusion/online_fusion_pass_reg istry.cc:35] DoOnlineFusionPass] do online fusion pass : DoReduceConcatFusionPass

[INFO] LITE(1490684,ffffa8360900,converter_lite):2024-11-26-18:36:46.930.193 [mindspore/lite/src/litert/pass/online_fusion/online_fusion_pass_reg istry.cc:35] DoOnlineFusionPass] do online fusion pass : DoSplitReduceConcatFusionPass

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:47.053.749 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:47.058.040 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffaa778020,converter_lite):2024-11-26-18:36:47.565.938 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:359] Fp32Inference] BC inference round: 1

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.109.556 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.109.679 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.113.532 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.113.611 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.113.648 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.113.680 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.134.416 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.134.556 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.148.684 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.148.815 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.148.873 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.148.946 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.226.147 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.226.465 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.285.093 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.285.416 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.285.488 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.285.742 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.305.498 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:213] operator()] tensor type is 43

[INFO] LITE(1490684,ffffa7b50900,converter_lite):2024-11-26-18:36:48.308.693 [mindspore/lite/tools/converter/quantizer/bias_correction_strategy.c c:255] operator()] tensor type is 43

3 根因分析

******此处由用户填写******

4 解决方案

******此处由用户填写******

包含文字方案和最终脚本代码

请将正确的脚本打包并上传附件

本帖最后由 匿名用户 于 2024/12/25 08:48:52 编辑

我要发帖子

我们使用cookie来确保您的高速浏览体验。继续浏览本站，即表示您同意我们使用cookie。查看详情