Skip to content

fix quantizedLinear layer feeds into src graph output #615

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
sfatimar merged 1 commit into from
Mar 18, 2025
Merged

fix quantizedLinear layer feeds into src graph output #615

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: 7 additions & 4 deletions onnxruntime/core/providers/openvino/qdq_transformations/qdq_stripping.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -346,7 +346,7 @@ static bool CheckQFeedsIntoQuantizedOutput(const NodeUnit& node_unit,
auto op_of_quantized_layer = node_unit.Outputs();
for (auto& itr : op_of_quantized_layer) {
auto it = graph_op_data_type.find(itr.node_arg.Name());
if (it != graph_op_data_type.end() && it->second == "tensor(uint8)") {
if (it != graph_op_data_type.end() && (it->second == "tensor(uint8)" || it->second == "tensor(uint16)")) {
return true;
}
}
Expand All @@ -369,6 +369,11 @@ static bool CheckQRuleSet(const NodeUnit& node_unit,
graph_op_data_type[src_graph.GetNodeArg(ops->Name())->Name()] = ops->Type()->data();
}

// check If any quantized node feeds into the src graph output
if (CheckQFeedsIntoQuantizedOutput(node_unit, std::move(graph_op_data_type))) {
return true;
}

// If UInt16 Q, don't keep it
if (GetQDQDataType(q_node) == DT_UINT16 || GetQDQDataType(q_node) == DT_INT16) {
reason = SkipReason::Int16QDQ;
Expand All @@ -381,9 +386,7 @@ static bool CheckQRuleSet(const NodeUnit& node_unit,
} else if (op_type == "Add") {
// Add keeps all Qs
return true;
} else if (CheckQFeedsIntoQuantizedOutput(node_unit, std::move(graph_op_data_type))) {
return true;
} else {
} else {
// Keep Q of an unsupported Op only if the target that succeeds it is a supported Op in this list
return IsNextTargetNodeOfQValid(q_node, &target_node, src_graph, {"Conv", "Add", "MatMul"}, false);
}
Expand Down
Loading