root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 03:14:35 +00:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

CUDA: use only 1 thread if fully offloaded (#2915)

Browse Source
This commit is contained in:
Johannes Gäßler 2023-09-21 10:43:53 +02:00 committed by GitHub
parent 7eb41179ed
commit 8185710a80
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 9 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
llama.cpp
View File

@ -3765,6 +3765,15 @@ static bool llama_eval_internal(
n_threads = std::min(4, n_threads); n_threads = std::min(4, n_threads);
} }
// If all tensors can be run on the GPU then using more than 1 thread is detrimental.
const bool full_offload_supported = model.arch == LLM_ARCH_LLAMA ||
model.arch == LLM_ARCH_BAICHUAN ||
model.arch == LLM_ARCH_FALCON;
const bool fully_offloaded = model.n_gpu_layers >= (int) hparams.n_layer + 3;
if (ggml_cpu_has_cublas() && full_offload_supported && fully_offloaded) {
n_threads = 1;
}
struct ggml_tensor * res = gf->nodes[gf->n_nodes - 1]; struct ggml_tensor * res = gf->nodes[gf->n_nodes - 1];
struct ggml_tensor * embeddings = gf->nodes[gf->n_nodes - 2]; struct ggml_tensor * embeddings = gf->nodes[gf->n_nodes - 2];