tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-28 11:08:19 -04:00
Code Issues Packages Projects Releases Wiki Activity

batched-bench : use rand tokens (#15398)

Browse Source
This commit is contained in:
Georgi Gerganov
2025-08-19 08:45:12 +03:00
committed by GitHub
parent f08c4c0d8d
commit f0d3c7405c
1 changed files with 10 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
tools/batched-bench/batched-bench.cpp
View File

@@ -57,6 +57,13 @@ int main(int argc, char ** argv) {
return 1;
}
const llama_vocab * vocab = llama_model_get_vocab(model);
const int32_t n_vocab = llama_vocab_n_tokens(vocab);
const auto get_token_rand = [n_vocab]() -> llama_token {
return std::rand() % n_vocab;
};
auto * mem = llama_get_memory(ctx);
const int32_t n_kv_max = llama_n_ctx(ctx);
@@ -93,7 +100,7 @@ int main(int argc, char ** argv) {
// warm up
{
for (int i = 0; i < 16; ++i) {
common_batch_add(batch, 0, i, { 0 }, false);
common_batch_add(batch, get_token_rand(), i, { 0 }, false);
}
if (!decode_helper(ctx, batch, ctx_params.n_batch)) {
@@ -127,7 +134,7 @@ int main(int argc, char ** argv) {
for (int j = 0; j < (is_pp_shared ? 1 : pl); ++j) {
for (int i = 0; i < pp; ++i) {
common_batch_add(batch, 0, i, { j }, i == pp - 1);
common_batch_add(batch, get_token_rand(), i, { j }, i == pp - 1);
}
}
@@ -154,7 +161,7 @@ int main(int argc, char ** argv) {
common_batch_clear(batch);
for (int j = 0; j < pl; ++j) {
common_batch_add(batch, 0, pp + i, { j }, true);
common_batch_add(batch, get_token_rand(), pp + i, { j }, true);
}
if (!decode_helper(ctx, batch, ctx_params.n_batch)) {