本当はllama3を解析しようかなと思ったんだけど、C言語なllama2.cを解析してみたいと思う。
関数の構成はこんな感じですね。
run.c 906 <0>main()
(null) |-<1>if()
run.c 891 |-<1>error_usage()
(null) | |-<2>fprintf()
(null) | +-<2>exit()
(null) |-<1>for()
(null) |-<1>strlen()
(null) |-<1>atof()
(null) |-<1>atoi()
(null) |-<1>time()
run.c 164 |-<1>build_transformer()
run.c 142 | |-<2>read_checkpoint()
(null) | | |-<3>fopen()
(null) | | |-<3>if()
(null) | | |-<3>fprintf()
(null) | | |-<3>exit()
(null) | | |-<3>fread()
(null) | | |-<3>sizeof()
(null) | | |-<3>abs()
(null) | | |-<3>fseek()
(null) | | |-<3>ftell()
(null) | | |-<3>fclose()
(null) | | |-<3>open()
(null) | | |-<3>mmap()
(null) | | |-<3>ssizeof()
run.c 111 | | +-<3>memory_map_weights()
run.c 77 | +-<2>malloc_run_state()
(null) | |-<3>calloc()
(null) | |-<3>sizeof()
(null) | |-<3>if()
(null) | |-<3>fprintf()
(null) | +-<3>exit()
run.c 385 |-<1>build_tokenizer()
(null) | |-<2>malloc()
(null) | |-<2>sizeof()
(null) | |-<2>for()
(null) | |-<2>fopen()
(null) | |-<2>if()
(null) | |-<2>fprintf()
(null) | |-<2>exit()
(null) | |-<2>fread()
(null) | +-<2>fclose()
run.c 667 |-<1>build_sampler()
(null) | |-<2>malloc()
(null) | +-<2>sizeof()
(null) |-<1>strcmp()
run.c 729 |-<1>generate()
(null) | |-<2>if()
(null) | |-<2>malloc()
(null) | |-<2>strlen()
(null) | |-<2>sizeof()
run.c 452 | |-<2>encode()
(null) | | |-<3>if()
(null) | | |-<3>fprintf()
(null) | | |-<3>exit()
(null) | | |-<3>malloc()
(null) | | |-<3>sizeof()
(null) | | |-<3>for()
(null) | | |-<3>qsort()
run.c 445 | | |-<3>str_lookup()
(null) | | | |-<4>bsearch()
(null) | | | +-<4>sizeof()
(null) | | |-<3>while()
(null) | | |-<3>sprintf()
(null) | | +-<3>free()
(null) | |-<2>fprintf()
(null) | |-<2>exit()
(null) | |-<2>while()
run.c 231 | |-<2>forward()
(null) | | |-<3>memcpy()
(null) | | |-<3>sizeof()
(null) | | |-<3>for()
run.c 182 | | |-<3>rmsnorm()
(null) | | | |-<4>for()
(null) | | | +-<4>sqrtf()
run.c 217 | | |-<3>matmul()
(null) | | | +-<4>for()
(null) | | |-<3>powf()
(null) | | |-<3>head_dim()
(null) | | |-<3>cosf()
(null) | | |-<3>sinf()
(null) | | |-<3>sqrtf()
run.c 197 | | |-<3>softmax()
(null) | | | |-<4>for()
(null) | | | |-<4>if()
(null) | | | +-<4>expf()
(null) | | |-<3>memset()
(null) | | |-<3>f()
(null) | | +-<3>expf()
run.c 691 | |-<2>sample()
(null) | | |-<3>if()
run.c 590 | | |-<3>sample_argmax()
(null) | | | |-<4>for()
(null) | | | +-<4>if()
(null) | | |-<3>for()
run.c 197 | | |-<3>softmax() ･･･
run.c 687 | | |-<3>random_f32()
(null) | | | |-<4>return()
run.c 680 | | | +-<4>random_u32()
(null) | | | +-<5>return()
run.c 603 | | |-<3>sample_mult()
(null) | | | |-<4>for()
(null) | | | +-<4>if()
run.c 624 | | +-<3>sample_topp()
(null) | | |-<4>for()
(null) | | |-<4>if()
(null) | | |-<4>qsort()
(null) | | +-<4>sizeof()
run.c 418 | |-<2>decode()
(null) | | |-<3>if()
(null) | | +-<3>sscanf()
run.c 431 | |-<2>safe_printf()
(null) | | |-<3>if()
(null) | | |-<3>isprint()
(null) | | |-<3>isspace()
(null) | | +-<3>printf()
(null) | |-<2>fflush()
run.c 719 | |-<2>time_in_ms()
(null) | | +-<3>clock_gettime()
(null) | |-<2>printf()
(null) | +-<2>free()
run.c 802 |-<1>chat()
(null) | |-<2>malloc()
(null) | |-<2>sizeof()
(null) | |-<2>while()
(null) | |-<2>if()
run.c 785 | |-<2>read_stdin()
(null) | | |-<3>printf()
(null) | | |-<3>if()
(null) | | |-<3>fgets()
(null) | | +-<3>strlen()
(null) | |-<2>strcpy()
(null) | |-<2>sprintf()
run.c 452 | |-<2>encode() ･･･
(null) | |-<2>printf()
run.c 231 | |-<2>forward() ･･･
run.c 691 | |-<2>sample() ･･･
run.c 418 | |-<2>decode() ･･･
run.c 431 | |-<2>safe_printf() ･･･
(null) | |-<2>fflush()
(null) | +-<2>free()
(null) |-<1>fprintf()
run.c 676 |-<1>free_sampler()
(null) | +-<2>free()
run.c 411 |-<1>free_tokenizer()
(null) | |-<2>for()
(null) | +-<2>free()
run.c 171 +-<1>free_transformer()
(null) |-<2>if()
(null) |-<2>munmap()
(null) |-<2>close()
run.c 98 +-<2>free_run_state()
(null) +-<3>free()
run.c 381 <0>compare_tokens()
(null) +-<1>strcmp()
run.c 616 <0>compare()
(null) +-<1>if()