tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 03:55:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

bench : handle decode errors (#13548)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-05-15 05:57:02 +03:00
committed by GitHub
parent aa48e373f2
commit b2838049cc
1 changed files with 39 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
tools/llama-bench/llama-bench.cpp
View File

@ -1736,7 +1736,7 @@ struct sql_printer : public printer {
} }
}; };
static void test_prompt(llama_context * ctx, int n_prompt, int n_batch, int n_threads) { static bool test_prompt(llama_context * ctx, int n_prompt, int n_batch, int n_threads) {
llama_set_n_threads(ctx, n_threads, n_threads); llama_set_n_threads(ctx, n_threads, n_threads);
const llama_model * model = llama_get_model(ctx); const llama_model * model = llama_get_model(ctx);
@ -1753,14 +1753,19 @@ static void test_prompt(llama_context * ctx, int n_prompt, int n_batch, int n_th
for (int i = 1; i < n_tokens; i++) { for (int i = 1; i < n_tokens; i++) {
tokens[i] = std::rand() % n_vocab; tokens[i] = std::rand() % n_vocab;
} }
llama_decode(ctx, llama_batch_get_one(tokens.data(), n_tokens)); int res = llama_decode(ctx, llama_batch_get_one(tokens.data(), n_tokens));
if (res != 0) {
fprintf(stderr, "%s: failed to decode prompt batch, res = %d\n", __func__, res);
return false;
}
n_processed += n_tokens; n_processed += n_tokens;
} }
llama_synchronize(ctx); llama_synchronize(ctx);
return true;
} }
static void test_gen(llama_context * ctx, int n_gen, int n_threads) { static bool test_gen(llama_context * ctx, int n_gen, int n_threads) {
llama_set_n_threads(ctx, n_threads, n_threads); llama_set_n_threads(ctx, n_threads, n_threads);
const llama_model * model = llama_get_model(ctx); const llama_model * model = llama_get_model(ctx);
@ -1770,10 +1775,15 @@ static void test_gen(llama_context * ctx, int n_gen, int n_threads) {
llama_token token = llama_vocab_get_add_bos(vocab) ? llama_vocab_bos(vocab) : std::rand() % n_vocab; llama_token token = llama_vocab_get_add_bos(vocab) ? llama_vocab_bos(vocab) : std::rand() % n_vocab;
for (int i = 0; i < n_gen; i++) { for (int i = 0; i < n_gen; i++) {
llama_decode(ctx, llama_batch_get_one(&token, 1)); int res = llama_decode(ctx, llama_batch_get_one(&token, 1));
if (res != 0) {
fprintf(stderr, "%s: failed to decode generation batch, res = %d\n", __func__, res);
return false;
}
llama_synchronize(ctx); llama_synchronize(ctx);
token = std::rand() % n_vocab; token = std::rand() % n_vocab;
} }
return true;
} }
static void llama_null_log_callback(enum ggml_log_level level, const char * text, void * user_data) { static void llama_null_log_callback(enum ggml_log_level level, const char * text, void * user_data) {
@ -1917,13 +1927,21 @@ int main(int argc, char ** argv) {
fprintf(stderr, "llama-bench: benchmark %d/%zu: warmup prompt run\n", params_idx, params_count); fprintf(stderr, "llama-bench: benchmark %d/%zu: warmup prompt run\n", params_idx, params_count);
} }
//test_prompt(ctx, std::min(t.n_batch, std::min(t.n_prompt, 32)), 0, t.n_batch, t.n_threads); //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, t.n_batch, t.n_threads); bool res = test_prompt(ctx, t.n_prompt, t.n_batch, t.n_threads);
if (!res) {
fprintf(stderr, "%s: error: failed to run prompt warmup\n", __func__);
exit(1);
}
} }
if (t.n_gen > 0) { if (t.n_gen > 0) {
if (params.progress) { if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%zu: warmup generation run\n", params_idx, params_count); fprintf(stderr, "llama-bench: benchmark %d/%zu: warmup generation run\n", params_idx, params_count);
} }
test_gen(ctx, 1, t.n_threads); bool res = test_gen(ctx, 1, t.n_threads);
if (!res) {
fprintf(stderr, "%s: error: failed to run gen warmup\n", __func__);
exit(1);
}
} }
for (int i = 0; i < params.reps; i++) { for (int i = 0; i < params.reps; i++) {
@ -1934,7 +1952,11 @@ int main(int argc, char ** argv) {
fprintf(stderr, "llama-bench: benchmark %d/%zu: depth run %d/%d\n", params_idx, params_count, fprintf(stderr, "llama-bench: benchmark %d/%zu: depth run %d/%d\n", params_idx, params_count,
i + 1, params.reps); i + 1, params.reps);
} }
test_prompt(ctx, t.n_depth, t.n_batch, t.n_threads); bool res = test_prompt(ctx, t.n_depth, t.n_batch, t.n_threads);
if (!res) {
fprintf(stderr, "%s: error: failed to run depth\n", __func__);
exit(1);
}
} }
uint64_t t_start = get_time_ns(); uint64_t t_start = get_time_ns();
@ -1944,14 +1966,22 @@ int main(int argc, char ** argv) {
fprintf(stderr, "llama-bench: benchmark %d/%zu: prompt run %d/%d\n", params_idx, params_count, fprintf(stderr, "llama-bench: benchmark %d/%zu: prompt run %d/%d\n", params_idx, params_count,
i + 1, params.reps); i + 1, params.reps);
} }
test_prompt(ctx, t.n_prompt, t.n_batch, t.n_threads); bool res = test_prompt(ctx, t.n_prompt, t.n_batch, t.n_threads);
if (!res) {
fprintf(stderr, "%s: error: failed to run prompt\n", __func__);
exit(1);
}
} }
if (t.n_gen > 0) { if (t.n_gen > 0) {
if (params.progress) { if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%zu: generation run %d/%d\n", params_idx, params_count, fprintf(stderr, "llama-bench: benchmark %d/%zu: generation run %d/%d\n", params_idx, params_count,
i + 1, params.reps); i + 1, params.reps);
} }
test_gen(ctx, t.n_gen, t.n_threads); bool res = test_gen(ctx, t.n_gen, t.n_threads);
if (!res) {
fprintf(stderr, "%s: error: failed to run gen\n", __func__);
exit(1);
}
} }
uint64_t t_ns = get_time_ns() - t_start; uint64_t t_ns = get_time_ns() - t_start;