sycl : offload of get_rows set to 0 (llama/10432)

This commit is contained in:
Alberto Cabrera Pérez 2024-11-29 12:38:45 +00:00 committed by Georgi Gerganov
parent d7c2a04bce
commit 964b154a2a

View File

@ -4493,7 +4493,7 @@ static bool ggml_backend_sycl_device_supports_buft(ggml_backend_dev_t dev, ggml_
static int64_t get_op_batch_size(const ggml_tensor * op) { static int64_t get_op_batch_size(const ggml_tensor * op) {
switch (op->op) { switch (op->op) {
case GGML_OP_GET_ROWS: case GGML_OP_GET_ROWS:
return op->ne[1]; // this will increse the speed of prefill in test return 0;
case GGML_OP_MUL_MAT: case GGML_OP_MUL_MAT:
return op->ne[1]; return op->ne[1];
case GGML_OP_MUL_MAT_ID: case GGML_OP_MUL_MAT_ID: