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

diff --git a/src/ggml-sycl/ggml-sycl.cpp b/src/ggml-sycl/ggml-sycl.cpp
index aabcdc22422fc104ded19bd91cd588ab38154ff5..808f74fa0b27e0f5c8733d3308c621dcfa16753b 100644 (file)
--- a/src/ggml-sycl/ggml-sycl.cpp
+++ b/src/ggml-sycl/ggml-sycl.cpp
@@ -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) {
     switch (op->op) {
         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:
             return op->ne[1];
         case GGML_OP_MUL_MAT_ID: