Skip to content

StaticAttentionIOManager: optional callback on logits from prefill #14336

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
merged 1 commit into from
Sep 16, 2025
Merged

StaticAttentionIOManager: optional callback on logits from prefill #14336

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
11 changes: 10 additions & 1 deletion examples/models/llama/runner/static_attention_io_manager.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -589,7 +589,9 @@ class StaticAttentionIOManager {
size_t prefill(
executorch::runtime::Span<TokenT> tokens,
executorch::runtime::Span<TokenT> input_buffer,
executorch::runtime::Method& method) {
executorch::runtime::Method& method,
std::function<void(executorch::runtime::Span<const float>)>
logits_callback = nullptr) {
ET_LOG(Info, "Prefilling at position %zu", input_pos_);
size_t input_len = input_buffer.size();
auto& masks = get_mask(input_buffer.size());
Expand All @@ -610,6 +612,13 @@ class StaticAttentionIOManager {
config_.k_cache_output_indices,
config_.v_cache_output_indices,
batch_len);
if (logits_callback) {
auto logits_tensor = method.get_output(0).toTensor();
auto* logits = logits_tensor.const_data_ptr<float>();
logits_callback(executorch::runtime::Span(
logits,
logits + batch_len * logits_tensor.size(logits_tensor.dim() - 1)));
}
}
return batch_len - 1;
}
Expand Down
Loading