refactor: register INT8 projection and attention fusion

This commit is contained in:
leejet committed 2026-10-31 20:01:41 +08:00
1 parent 1d2d31ffb6
commit ab232a474b
8 files changed
+491 -19

No files matched your search

+2
View File
@@ -12,6 +12,7 @@
void sd_ggml_register_cuda_rope(ggml_backend_dev_t device);
#endif
#ifdef SD_CUDA_CUTE_ATTENTION
void sd_ggml_register_cuda_i8_projection(ggml_backend_dev_t device);
void sd_ggml_register_cuda_attention(ggml_backend_dev_t device);
#endif
@@ -180,6 +181,7 @@ void sd_ggml_register_extensions(ggml_backend_t backend) {
sd_ggml_register_cuda_rope(device);
#endif
#ifdef SD_CUDA_CUTE_ATTENTION
sd_ggml_register_cuda_i8_projection(device);
sd_ggml_register_cuda_attention(device);
#endif
return;