vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-06 05:09:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix: add warm-up in profiling to prevent init delay

Browse source
This commit is contained in:
Li, Zonghang 2025-06-03 17:10:09 +04:00
parent b30f749e5e
commit 1b3b6a506f
2 changed files with 9 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
common/profiler.cpp
View file

@ -439,7 +439,7 @@ float device_inp_embd_delay(struct llama_model * model, enum ggml_type src0t, in
}
// warm-up
// ggml_backend_graph_compute(backend, gf);
ggml_backend_graph_compute(backend, gf);
const int64_t t_start = ggml_time_us();
ggml_backend_graph_compute(backend, gf);
@ -1288,6 +1288,9 @@ static float device_mem_copy(struct llama_model * model, enum profiler_backend_t
ggml_backend_cpu_set_n_threads(backend, n_threads);
}
// warm-up
ggml_backend_graph_compute(backend, gf);
const int64_t t_start = ggml_time_us();
ggml_backend_graph_compute(backend, gf);
const int64_t t_end = ggml_time_us();

10
src/llama.cpp
View file

@ -18211,7 +18211,7 @@ static int llama_decode_internal(
[&]{ llama_kv_cache_clear (&lctx); },
[&]{ llama_send_kv_cache_clear (&lctx); },
is_last_dev)) {
LLAMA_LOG_INFO("%s: received signal kv_cache_clear\n", __func__);
LLAMA_LOG_DEBUG("%s: received signal kv_cache_clear\n", __func__);
return -1;
}
@ -18219,7 +18219,7 @@ static int llama_decode_internal(
[&]{ llama_kv_cache_seq_rm (&lctx, meta.rm_seq_id, meta.rm_p0, meta.rm_p1); },
[&]{ llama_send_kv_cache_seq_rm (&lctx, meta.rm_seq_id, meta.rm_p0, meta.rm_p1); },
is_last_dev)) {
LLAMA_LOG_INFO("%s: received signal kv_cache_seq_rm\n", __func__);
LLAMA_LOG_DEBUG("%s: received signal kv_cache_seq_rm\n", __func__);
return -1;
}
@ -18227,7 +18227,7 @@ static int llama_decode_internal(
[&]{ llama_kv_cache_seq_add (&lctx, meta.add_seq_id, meta.add_p0, meta.add_p1, meta.add_delta); },
[&]{ llama_send_kv_cache_seq_add(&lctx, meta.add_seq_id, meta.add_p0, meta.add_p1, meta.add_delta); },
is_last_dev)) {
LLAMA_LOG_INFO("%s: received signal kv_cache_seq_add\n", __func__);
LLAMA_LOG_DEBUG("%s: received signal kv_cache_seq_add\n", __func__);
return -1;
}
@ -18235,7 +18235,7 @@ static int llama_decode_internal(
[&]{ llama_kv_cache_seq_cp (&lctx, meta.cp_src_seq_id, meta.cp_dst_seq_id, meta.cp_p0, meta.cp_p1); },
[&]{ llama_send_kv_cache_seq_cp (&lctx, meta.cp_src_seq_id, meta.cp_dst_seq_id, meta.cp_p0, meta.cp_p1); },
is_last_dev)) {
LLAMA_LOG_INFO("%s: received signal kv_cache_seq_cp\n", __func__);
LLAMA_LOG_DEBUG("%s: received signal kv_cache_seq_cp\n", __func__);
return -1;
}
@ -18243,7 +18243,7 @@ static int llama_decode_internal(
[&]{ llama_kv_cache_seq_div (&lctx, meta.div_seq_id, meta.div_p0, meta.div_p1, meta.div_factor); },
[&]{ llama_send_kv_cache_seq_div(&lctx, meta.div_seq_id, meta.div_p0, meta.div_p1, meta.div_factor); },
is_last_dev)) {
LLAMA_LOG_INFO("%s: received signal kv_cache_seq_div\n", __func__);
LLAMA_LOG_DEBUG("%s: received signal kv_cache_seq_div\n", __func__);
return -1;
}
}