Prevent offloading of more than 33 layers
This commit is contained in:
parent
27d0c11897
commit
51b3b56c08
1 changed files with 8 additions and 0 deletions
|
@ -3020,6 +3020,14 @@ static void llm_load_tensors(
|
|||
ggml_backend_type backend_norm;
|
||||
ggml_backend_type backend_output;
|
||||
|
||||
// Don't allow for offloading of more than 33 layers.
|
||||
// Offloading 34 layers causes model to respond with letter 'E'
|
||||
// Offloading 35 layers doesn't work because of missing cuda implementation for rope:
|
||||
// GGML_ASSERT: ggml-cuda.cu:6402: ne00 == n_dims && "ne00 != n_dims is not implemented for CUDA yet"
|
||||
if (n_gpu_layers > 33) {
|
||||
n_gpu_layers = 33;
|
||||
}
|
||||
|
||||
if (n_gpu_layers > int(n_layer)) {
|
||||
// norm is not performance relevant on its own but keeping it in VRAM reduces data copying
|
||||
// on Windows however this is detrimental unless everything is on the GPU
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue