vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : llama_perf + option to disable timings during decode (#9355)

Browse source 
* llama : llama_perf + option to disable timings during decode

ggml-ci

* common : add llama_arg

* Update src/llama.cpp

Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>

* perf : separate functions in the API

ggml-ci

* perf : safer pointer handling + naming update

ggml-ci

* minor : better local var name

* perf : abort on invalid sampler pointer

ggml-ci

---------

Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
This commit is contained in:
Georgi Gerganov 2024-09-13 09:53:38 +03:00 • committed by GitHub
parent bd35cb0ae3
commit 0abc6a2c25
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
23 changed files with 135 additions and 91 deletions
Download patch file Download diff file Expand all files Collapse all files

34
src/llama-sampling.cpp
View file

@ -1669,3 +1669,37 @@ uint32_t llama_sampler_get_seed(const struct llama_sampler * smpl) {
return LLAMA_DEFAULT_SEED;
}
// perf
struct llama_perf_sampler_data llama_perf_sampler(const struct llama_sampler * chain) {
struct llama_perf_sampler_data data = {};
if (chain == nullptr || chain->iface != &llama_sampler_chain_i) {
GGML_ABORT("%s: invalid sampler passed - requires a sampler created with llama_sampler_chain_init()\n", __func__);
}
const auto * ctx = (const struct llama_sampler_chain *) chain->ctx;
data.t_sample_ms = 1e-3 * ctx->t_sample_us;
data.n_sample = std::max(0, ctx->n_sample);
return data;
}
void llama_perf_sampler_print(const struct llama_sampler * chain) {
const auto data = llama_perf_sampler(chain);
LLAMA_LOG_INFO("%s: sampling time = %10.2f ms / %5d runs (%8.2f ms per token, %8.2f tokens per second)\n",
__func__, data.t_sample_ms, data.n_sample, data.t_sample_ms / data.n_sample, 1e3 / data.t_sample_ms * data.n_sample);
}
void llama_perf_sampler_reset(struct llama_sampler * chain) {
if (chain == nullptr || chain->iface != &llama_sampler_chain_i) {
GGML_ABORT("%s: invalid sampler passed - requires a sampler created with llama_sampler_chain_init()\n", __func__);
}
auto * ctx = (struct llama_sampler_chain *) chain->ctx;
ctx->t_sample_us = ctx->n_sample = 0;
}