About issus

Results 13 comments of About issus

About the "lib_app", when i run the build.sh, have a issus: usr/bin/ld: 当搜索用于 /home/lib_app/tflib/lib/libtensorflow-lite.so 时跳过不兼容的 -ltensorflow-lite /usr/bin/ld: 当搜索用于 /home/lib_app/tflib/lib/libtensorflow-lite.a 时跳过不兼容的 -ltensorflow-lite /usr/bin/ld: 找不到 -ltensorflow-lite Thanks

是否是在纯c环境下无法运行ncnn

我将src的cmakelists修改成: if(NCNN_TARGET_ARCH STREQUAL "riscv") target_compile_options(ncnn PRIVATE -march=rv32imafdcpzpsfoperand_xtheade -mabi=ilp32d -mcmodel=medlow -g2 -Os -DSHL_BUILD_REF -DSHL_BUILD_GREF -DSHL_BUILD_E907 -DSHL_BUILD_RTOS) endif() 是可以得到elf32了,但是会出现错误: [100%] Linking CXX executable benchncnn /home/Xuantie-900-gcc-elf-newlib-x86_64-V2.0.5/bin/../lib/gcc/riscv64-unknown-elf/10.2.0/../../../../riscv64-unknown-elf/bin/ld: ../src/libncnn.a(allocator.cpp.obj): ABI is incompatible with that of...

如果注释掉 add_subdirectory(benchmark)没报错,但是得到的库文件会特别大,30+M

And the output results of each run are different.

st->rnn.model->vad_gru_state = 4282656 st->rnn.model->noise_gru_state = 4282768 st->rnn.model->denoise_gru_size = 4282976

我想问下,和tflite使用一样的量化方法,得到的参数差异也不大,同样的输入,但是为什么计算出来的结果确是完全不一样

感谢您的回复,我看对于给定输出是tensor_output_dim = {1,256,2},输出的nnom_output_data好像是一维的数组512,它们的对应关系是什么样的,是前256为[0][256][0]维与后256为[0][256][1]组成还是交叉组成512的一维数组

谢谢,假设我有输入数据为实部128,虚部128,那我是不是可以之间将它们放在一起组成[256]个数组作为输入呢,是之间实部+虚部进行组合,还是应该实虚实虚这样交叉放入

Hi: I used a instead of b in my model, and when generating the. h file, it gave me "model. hook (reshappe_s (&tf. reshappe_12_config), layer [1];", but there was no...