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

tests : disable grad / opt + minor naming changes

Browse source
This commit is contained in:
Georgi Gerganov 2023-07-06 20:30:40 +03:00
parent 4646cc2cf1
commit 8e1f0b6865
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 7 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

9
llama.cpp
View file

@ -1268,7 +1268,7 @@ static bool llama_eval_internal(
const float * embd, const float * embd,
const int n_tokens, const int n_tokens,
const int n_past, const int n_past,
const int n_threads, int n_threads,
const char * cgraph_fname) { const char * cgraph_fname) {
LLAMA_ASSERT((!tokens && embd) || (tokens && !embd)); LLAMA_ASSERT((!tokens && embd) || (tokens && !embd));
@ -1309,10 +1309,11 @@ static bool llama_eval_internal(
struct ggml_context * ctx0 = ggml_init(params); struct ggml_context * ctx0 = ggml_init(params);
ggml_cgraph gf = {};
// for big prompts, if BLAS is enabled, it is better to use only one thread // for big prompts, if BLAS is enabled, it is better to use only one thread
// otherwise, the threads are spin-lock waiting for the BLAS calls and are degrading the performance // otherwise, the threads are spin-lock waiting for the BLAS calls and are degrading the performance
ggml_cgraph gf = {}; n_threads = N >= 32 && ggml_cpu_has_blas() && !ggml_cpu_has_gpublas() ? 1 : n_threads;
const int actual_n_threads = N >= 32 && ggml_cpu_has_blas() && !ggml_cpu_has_gpublas() ? 1 : n_threads;
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * inpL; struct ggml_tensor * inpL;
@ -1622,7 +1623,7 @@ static bool llama_eval_internal(
#endif #endif
if (call_ggml_graph_compute) { if (call_ggml_graph_compute) {
ggml_cplan pf = ggml_graph_plan(&gf, actual_n_threads); ggml_cplan pf = ggml_graph_plan(&gf, n_threads);
if (pf.work_size > 0) { if (pf.work_size > 0) {
lctx.work_buffer.resize(pf.work_size); lctx.work_buffer.resize(pf.work_size);
pf.work_data = lctx.work_buffer.data(); pf.work_data = lctx.work_buffer.data();

4
tests/CMakeLists.txt
View file

@ -10,5 +10,5 @@ llama_add_test(test-quantize-fns.cpp)
llama_add_test(test-quantize-perf.cpp) llama_add_test(test-quantize-perf.cpp)
llama_add_test(test-sampling.cpp) llama_add_test(test-sampling.cpp)
llama_add_test(test-tokenizer-0.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab.bin) llama_add_test(test-tokenizer-0.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab.bin)
llama_add_test(test-grad0.c) # SLOW # llama_add_test(test-grad0.c) # SLOW
llama_add_test(test-opt.c) # SLOW # llama_add_test(test-opt.c) # SLOW