llama : add guard for K/V rotation input when buffer is unallocated (#25215)

llm_graph_input_attn_kv::set_input and llm_graph_input_attn_kv_iswa::set_input
call set_input_k_rot / set_input_v_rot whenever the rotation tensor pointer is
non-null, but the tensor's buffer can be unallocated (NULL) when a graph only
stores K/V without attending -- e.g. DFlash speculative decoding's KV-injection
pass. set_input_k_rot then calls ggml_backend_buffer_is_host() on a NULL buffer
and aborts with GGML_ASSERT(buffer).

Guard the four k_rot/v_rot inputs with the same "&& ->buffer" check that the
adjacent kq_mask inputs already use in these two functions. When the buffer is
unallocated there is no data to upload, so skipping is correct.

Fixes #25191

Signed-off-by: liminfei-amd <91481003+liminfei-amd@users.noreply.github.com>
This commit is contained in:
liminfei-amd
2026-07-05 04:37:38 +08:00
committed by GitHub
parent 665892536d
commit a4107133a6

View File

@@ -494,11 +494,11 @@ void llm_graph_input_attn_kv::set_input(const llama_ubatch * ubatch) {
mctx->set_input_kq_mask(self_kq_mask, ubatch, cparams.causal_attn);
}
if (self_k_rot) {
if (self_k_rot && self_k_rot->buffer) {
mctx->set_input_k_rot(self_k_rot);
}
if (self_v_rot) {
if (self_v_rot && self_v_rot->buffer) {
mctx->set_input_v_rot(self_v_rot);
}
}
@@ -592,19 +592,19 @@ void llm_graph_input_attn_kv_iswa::set_input(const llama_ubatch * ubatch) {
mctx->get_swa()->set_input_kq_mask(self_kq_mask_swa, ubatch, cparams.causal_attn);
}
if (self_k_rot) {
if (self_k_rot && self_k_rot->buffer) {
mctx->get_base()->set_input_k_rot(self_k_rot);
}
if (self_v_rot) {
if (self_v_rot && self_v_rot->buffer) {
mctx->get_base()->set_input_v_rot(self_v_rot);
}
if (self_k_rot_swa) {
if (self_k_rot_swa && self_k_rot_swa->buffer) {
mctx->get_swa()->set_input_k_rot(self_k_rot_swa);
}
if (self_v_rot_swa) {
if (self_v_rot_swa && self_v_rot_swa->buffer) {
mctx->get_swa()->set_input_v_rot(self_v_rot_swa);
}
}