ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-28 02:11:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fused soft cap and SIMD-ified GeLU (#9)

Browse Source 
* Softcap: WIP

Fuses scale + tanh + scale as used for softcaping in some
models.

Just CPU for now. ~1.4% for PP-512 on Gemma2-9b, no effect on TG.

Somewhat surprisingly the improvement does not increase as I
go to longer contexts. Gemma2 does softcap on K*Q, which grows
quadratically with context length, so I would have thought
the benefit from fusing scale, tanh, scale would increase.
But no, no luck.

* softcap: CUDA

* softcap: CUDA

~1% speedup for Gemma2-9b

* softcap: Metal and NEON

About 1% speedup.

* Simdified gelu

Gives ~1% speedup for Gemma2-9b prompt processing on AVX512/AVX2.
It looks like the gelu operation is memory bound on my CPU's
after SIMD-ifying it. By not using the 128 kb gelu lookup table
we gain a small advantage.
On the M2-Max the lookup table is slightly faster than the SIMD
version, so left the lookup table for ARM_NEON.

* softcap, tanh: avoid NaNs for large arguments (AVX2, AVX512)

Not that I have encountered this in practice, but just to be sure.
This does it for AVX512 and AVX2, still need a guard for ARM_NEON.

* llama-bench: add ability to turn off warmup runs

So we don't need to wait forever on, e.g., benchmarks involving
long contexts.

* softcap, tanh: avoid NaNs for large arguments (NEON)

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2024-08-20 17:15:47 +03:00
committed by GitHub
parent 38dcba95fe
commit 8a10467990
9 changed files with 437 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
examples/llama-bench/llama-bench.cpp
View File

@@ -237,6 +237,7 @@ struct cmd_params {
ggml_numa_strategy numa;
int reps;
bool verbose;
bool warmup;
output_formats output_format;
output_formats output_format_stderr;
};
@@ -263,6 +264,7 @@ static const cmd_params cmd_params_defaults = {
/* numa */ GGML_NUMA_STRATEGY_DISABLED,
/* reps */ 5,
/* verbose */ false,
/* warmup */ true,
/* output_format */ MARKDOWN,
/* output_format_stderr */ NONE,
};
@@ -295,6 +297,7 @@ static void print_usage(int /* argc */, char ** argv) {
printf(" -o, --output <csv|json|md|sql> (default: %s)\n", output_format_str(cmd_params_defaults.output_format));
printf(" -oe, --output-err <csv|json|md|sql> (default: %s)\n", output_format_str(cmd_params_defaults.output_format_stderr));
printf(" -v, --verbose (default: %s)\n", cmd_params_defaults.verbose ? "1" : "0");
printf(" -w, --warmup <0|1> (default: %s)\n", cmd_params_defaults.warmup ? "1" : "0");
printf("\n");
printf("Multiple values can be given for each parameter by separating them with ',' or by specifying the parameter multiple times.\n");
}
@@ -338,6 +341,7 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
params.output_format_stderr = cmd_params_defaults.output_format_stderr;
params.reps = cmd_params_defaults.reps;
params.numa = cmd_params_defaults.numa;
params.warmup = cmd_params_defaults.warmup;
for (int i = 1; i < argc; i++) {
arg = argv[i];
@@ -555,6 +559,12 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
invalid_param = !output_format_from_str(argv[i], params.output_format_stderr);
} else if (arg == "-v" || arg == "--verbose") {
params.verbose = true;
} else if (arg == "-w" || arg == "--warmup") {
if (++i >= argc) {
invalid_param = true;
break;
}
params.warmup = std::stoi(argv[i]);
} else {
invalid_param = true;
break;
@@ -1429,12 +1439,14 @@ int main(int argc, char ** argv) {
llama_kv_cache_clear(ctx);
// warmup run
if (t.n_prompt > 0) {
//test_prompt(ctx, std::min(t.n_batch, std::min(t.n_prompt, 32)), 0, t.n_batch, t.n_threads);
test_prompt(ctx, t.n_prompt, 0, t.n_batch, t.n_threads);
}
if (t.n_gen > 0) {
test_gen(ctx, 1, 0, t.n_threads);
if (params.warmup) {
if (t.n_prompt > 0) {
//test_prompt(ctx, std::min(t.n_batch, std::min(t.n_prompt, 32)), 0, t.n_batch, t.n_threads);
test_prompt(ctx, t.n_prompt, 0, t.n_batch, t.n_threads);
}
if (t.n_gen > 0) {
test_gen(ctx, 1, 0, t.n_threads);
}
}
for (int i = 0; i < params.reps; i++) {