Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-05 18:44:51 +01:00
Code Issues Packages Projects Releases Wiki Activity

mpi : minor

Browse Source
This commit is contained in:
Georgi Gerganov 2023-07-09 18:38:32 +03:00
parent beadbf3380
commit 9da9d26c70
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
2 changed files with 6 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ggml-mpi.c
View File

@ -6,7 +6,6 @@
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#define MIN(a, b) ((a) < (b) ? (a) : (b))
@ -168,7 +167,8 @@ void ggml_mpi_graph_compute(
return;
}
// attach the input data to the first layer for this node
// attach the input data to all nodes that need it
// TODO: not great - should be able to do this without modifying the compute graph (see next TODO below)
for (int i = idx_l0; i < idx_l1; i++) {
if (gf->nodes[i]->src0 == gf->nodes[idx_l0]) {
gf->nodes[i]->src0 = inp0;

4
llama.cpp
View File

@ -1342,6 +1342,10 @@ static bool llama_eval_internal(
inpL = ggml_get_rows(ctx0, model.tok_embeddings, inp_tokens);
} else {
#ifdef GGML_USE_MPI
GGML_ASSERT(false && "not implemented");
#endif
inpL = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, N);
memcpy(inpL->data, embd, N * n_embd * ggml_element_size(inpL));
}