Skip to content

ggml-openvino: accept flash attention on layers without RoPE #372

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
cavusmustafa merged 1 commit into from
Nov 3, 2026
+6 −1
Merged

ggml-openvino: accept flash attention on layers without RoPE #372

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion ggml/src/ggml-openvino/ggml-openvino.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1330,6 +1330,7 @@ static bool is_supported_flash_attn_pattern(const ggml_tensor * op) {
// PERMUTE -> VIEW -> base (view_src==nullptr) (llama KV-cache path)
// PERMUTE -> RESHAPE -> base (view_src==nullptr) (whisper Q)
// VIEW -> base (view_src==nullptr) (whisper K/V from kv_pad)
// PERMUTE -> VIEW -> RESHAPE -> base (Q of a layer without RoPE, e.g. SmolLM3)
for (int i = 0; i < 3; i++) {
const ggml_tensor * src = op->src[i];
if (src->op == GGML_OP_PERMUTE) {
Expand All @@ -1339,7 +1340,11 @@ static bool is_supported_flash_attn_pattern(const ggml_tensor * op) {
if (src->src[0]->op != GGML_OP_VIEW && src->src[0]->op != GGML_OP_RESHAPE) {
return false;
}
if (src->src[0]->src[0] == nullptr || src->src[0]->src[0]->view_src != nullptr) {
const ggml_tensor * base = src->src[0]->src[0];
if (base != nullptr && src->src[0]->op == GGML_OP_VIEW && base->op == GGML_OP_RESHAPE) {
base = base->src[0];
}
if (base == nullptr || base->view_src != nullptr) {
return false;
}
} else if (src->op == GGML_OP_VIEW) {
Expand Down