For the second question, I added #include <cmath> at the beginning of Gelu.cc, which fixed the bug.
For the first question, I tried to use NNAPCK to fix the acceleration problem. However, I could not locate the CMakeList.txt in TinyChatEngine when I built the whole NNAPCK. Lol. I am not sure this is the correct way.
Hi team,
Thanks for your amazing work.
I met a problem while deploying models in my Google Pixel 8 Pro(ARM).
`../kernels/neon/matmul_neon_fp32.cc:6:10: fatal error: Accelerate/Accelerate.h: No such file or directory
6 | #include <Accelerate/Accelerate.h>
| ^
~~~~~~~~compilation terminated.
make: *** [Makefile:181: build/transformer/../kernels/neon/matmul_neon_fp32.o] Error 1
make: *** Waiting for unfinished jobs....
../kernels/neon/matmul_neon_int8_int4.cc:7:10: fatal error: Accelerate/Accelerate.h: No such file or directory
7 | #include <Accelerate/Accelerate.h>
| ^
~~~~~~~~compilation terminated.
make: *** [Makefile:181: build/transformer/../kernels/neon/matmul_neon_int8_int4.o] Error 1
src/ops/Gelu.cc: In function ‘float Gelu_imp(float)’:
src/ops/Gelu.cc:10:31: error: ‘tanhf’ was not declared in this scope
10 | return 0.5f x (1.0f + tanhf(SQRT_2_OVER_PI x (1.0f + GELU_COEF_A x x)));
| ^~~~~
src/ops/Gelu.cc: In function ‘float Gelu_quick_imp(float)’:
src/ops/Gelu.cc:20:32: error: ‘expf’ was not declared in this scope
20 | return x (1.0f / (1.0f + expf(GELU_QUICK_COEF x)));
| ^~~~
make: *** [Makefile:181: build/transformer/src/ops/Gelu.o] Error 1`
Could you help me fix the bugs? Thanks
Cheers, Jeremy