root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-05 00:04:36 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

Revert the prompt processing on gpu for now.

Browse Source 
Fixes issues #1580 and #1581
This commit is contained in:
Adam Treat 2023-10-27 18:32:51 -04:00 committed by cebtenzzre
parent e006d377dd
commit a5eb001eab
1 changed files with 10 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
llama.cpp
View File

@ -3870,11 +3870,19 @@ static bool llama_eval_internal(
ggml_graph_compute_helper(lctx.work_buffer, gf, n_threads);
}
#elif defined(GGML_USE_KOMPUTE)
if (lctx.ctx_kompute) {
if (lctx.ctx_kompute && N == 1) {
ggml_vk_graph_compute(lctx.ctx_kompute, gf);
ggml_vk_d2h_tensor(lctx.ctx_kompute, res);
} else {
if (lctx.ctx_kompute) {
ggml_vk_d2h_tensor(lctx.ctx_kompute, kv_self.k);
ggml_vk_d2h_tensor(lctx.ctx_kompute, kv_self.v);
}
ggml_graph_compute_helper(lctx.work_buffer, gf, n_threads);
if (lctx.ctx_kompute) {
ggml_vk_h2d_tensor(lctx.ctx_kompute, kv_self.k);
ggml_vk_h2d_tensor(lctx.ctx_kompute, kv_self.v);
}
}
#else
ggml_graph_compute_helper(lctx.work_buffer, gf, n_threads);