gesanqiu/Chinese_MobileBert_on_SNPE

运行时遇到了错误,请求帮助

Tom-plus opened this issue · 1 comments

1.运行时报错:./test-text2vec
[2024-03-07 18:55:42.359] [info] Set log level to info.
2024-03-07 18:55:42.382918 <thread 53886> [info] [/home/song/Chinese_MobileBert_on_SNPE-main/snpe_bert/SNPETask.cpp:65] [SNPETask] Using SNPE: 2.19.0.240124133650_81096
2024-03-07 18:55:42.564300 <thread 53886> [error] [/home/song/Chinese_MobileBert_on_SNPE-main/snpe_bert/SNPETask.cpp:128] [init] SNPE build failed: error_code=7005; error_message= QnnGraph_addNode() failed for layer: Unsqueeze_48; QNN_GRAPH_ERROR_INVALID_OP_CONFIG: Some elements in the op config data are invalid; error_component=QNN; line_no=138; thread_id=139691552315520
2024-03-07 18:55:42.564415 <thread 53886> [error] [/home/song/Chinese_MobileBert_on_SNPE-main/snpe_bert/Bert.cpp:43] [Init] Can't init snpetask instance.
2024-03-07 18:55:42.565969 <thread 53886> [error] [/home/song/Chinese_MobileBert_on_SNPE-main/snpe_bert/SNPETask.cpp:273] [getInputTensor] The getInputTensor() needs to be called after SNPETask is initialized!
2024-03-07 18:55:42.566068 <thread 53886> [error] [/home/song/Chinese_MobileBert_on_SNPE-main/snpe_bert/SNPETask.cpp:273] [getInputTensor] The getInputTensor() needs to be called after SNPETask is initialized!
Segmentation fault (core dumped)
2.是否漏传了一个文件:mobilebert_gather99_weights.bin

1.用Netron检查模型Unsqueeze_48层,对照QNN/SNPE关于unsqueeze的描述,感觉是维度的问题;
2.有相关的报错吗,我应该没用到过这个weight,文档有描述gather层是怎么改的。