Sync from upstream llama.cpp repository

2026-01-16 10:43:34 +08:00
parent 3bc369a6f7
commit f4ae4cc7da
2053 changed files with 956010 additions and 1 deletions
--- a/ggml/src/ggml-cuda/fattn.cuh
+++ b/ggml/src/ggml-cuda/fattn.cuh
@@ -0,0 +1,5 @@
+#include "common.cuh"
+
+void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+
+bool ggml_cuda_flash_attn_ext_supported(int device, const ggml_tensor * dst);