Skip to content

Commit dbf17f0

Browse files
committed
kompute : disable LLAMA_SPLIT_LAYER after ggml-org#5321
Signed-off-by: Jared Van Bortel <[email protected]>
1 parent 4a06b01 commit dbf17f0

File tree

1 file changed

+4
-1
lines changed

1 file changed

+4
-1
lines changed

llama.cpp

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -3763,6 +3763,7 @@ static bool llm_load_tensors(
37633763
model.buft_layer[i] = llama_default_buffer_type_cpu(true);
37643764
}
37653765

3766+
#ifndef GGML_USE_KOMPUTE
37663767
if (split_mode == LLAMA_SPLIT_MODE_LAYER) {
37673768
// calculate the split points
37683769
int device_count = llama_get_device_count();
@@ -3800,7 +3801,9 @@ static bool llm_load_tensors(
38003801
} else {
38013802
model.buft_output = llama_default_buffer_type_cpu(true);
38023803
}
3803-
} else {
3804+
} else
3805+
#endif
3806+
{
38043807
ggml_backend_buffer_type_t split_buft;
38053808
if (split_mode == LLAMA_SPLIT_MODE_ROW) {
38063809
split_buft = llama_default_buffer_type_split(main_gpu, tensor_split);

0 commit comments

Comments
 (0)