Skip to content

bugfix: redirect draft-extend placeholder KV write to padding block. #1944

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

Open
weizhehuang0827 wants to merge 1 commit into
base: main
Choose a base branch
from
+12 −2
Open

bugfix: redirect draft-extend placeholder KV write to padding block. #1944

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
14 changes: 12 additions & 2 deletions xllm/core/runtime/mtp_worker_impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1526,11 +1526,16 @@ void MTPWorkerImpl::prepare_draft_extend_inputs(
if (use_chunked_prefill) {
int32_t prev_token_id = state.prev_token_id;
torch::Tensor prev_embedding = state.prev_embedding;
if (prev_token_id < 0) {
const bool prev_is_placeholder = prev_token_id < 0;
if (prev_is_placeholder) {
prev_token_id = current_token_id >= 0 ? current_token_id : 0;
prev_embedding = torch::Tensor();
}
add_row(prev_token_id, /*position_offset=*/-1, prev_embedding);
if (prev_is_placeholder) {
// Redirect to padding block 0 to avoid overwriting correct KV cache.
buf.out_new_cache_slots.back() = 0;
}
add_row(state.token_id, /*position_offset=*/0, state.embedding);
specBuilder::append_seq_len_by_layout(buf.out_q_seq_lens, 2);
const int32_t kv_len = specBuilder::calc_kv_len(
Expand All @@ -1547,11 +1552,16 @@ void MTPWorkerImpl::prepare_draft_extend_inputs(
int32_t prev_token_id = state.prev_token_id;
int32_t prev_position_offset = -1;
torch::Tensor prev_embedding = state.prev_embedding;
if (prev_token_id < 0) {
const bool prev_is_placeholder = prev_token_id < 0;
if (prev_is_placeholder) {
prev_token_id = state.token_id;
prev_embedding = torch::Tensor();
}
add_row(prev_token_id, prev_position_offset, prev_embedding);
if (prev_is_placeholder) {
// Redirect to padding block 0 to avoid overwriting correct KV cache.
buf.out_new_cache_slots.back() = 0;
}
}

selected_row_idx.emplace_back(
Expand Down
Loading