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

llama : remove obsolete code

Browse source
This commit is contained in:
Stanisław Szymczyk 2024-07-02 11:22:58 +02:00
parent 78675f35ee
commit 1d1cb01b01
1 changed files with 0 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

11
src/llama.cpp
View file

@ -13767,17 +13767,6 @@ static int llama_encode_internal(
GGML_ASSERT(strcmp(embd->name, "result_norm") == 0);
// 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
// TODO: this is mostly important for Apple Silicon where CBLAS is still performing very well
// we still need some threads to process all non-mul_mat ops, but not too much to avoid interfering
// with the BLAS calls. need a better solution
// MoE Special Case: This logic applies when hparams.n_expert == 0, i.e. the model is NOT an MoE model. When an MoE is
// being processed then Accelerate/BLAS will not be involved, so capping would limit performance.
if (n_tokens >= 32 && hparams.n_expert == 0 && ggml_cpu_has_blas() && !ggml_cpu_has_gpublas()) {
n_threads = std::min(4, n_threads);
}
ggml_backend_sched_alloc_graph(lctx.sched, gf);
llama_set_inputs(lctx, u_batch);