Skip to content

[5477976] Fix: issue removing Q/DQ nodes around custom ops with constant inputs #296

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 5, 2025
Merged

[5477976] Fix: issue removing Q/DQ nodes around custom ops with constant inputs #296

Show file tree
Hide file tree
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
13 changes: 9 additions & 4 deletions modelopt/onnx/quantization/qdq_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -790,8 +790,10 @@ def remove_input_dq_and_output_q(
if cons_idx in quantizable_custom_ops[consumer.op_type]["inp"]:
consumer.input[cons_idx] = q_node.output[0]
else:
q_node_prev = tensor_producers[q_node.input[0]]
consumer.input[cons_idx] = q_node_prev.output[0]
q_node_prev = tensor_producers.get(q_node.input[0], None)
consumer.input[cons_idx] = (
q_node_prev.output[0] if q_node_prev else q_node.input[0]
)
break

# Track DequantizeLinear node indices for cleanup
Expand Down Expand Up @@ -828,8 +830,11 @@ def remove_input_dq_and_output_q(
if quantizable_custom_ops[producer.op_type]["out"]:
dq_node[0].input[0] = producer.output[0]
else:
dq_node_next = tensor_consumers[dq_node[0].output[0]]
dq_node_next[0].input[0] = producer.output[0]
dq_node_next = tensor_consumers.get(dq_node[0].output[0], None)
if dq_node_next:
dq_node_next[0].input[0] = producer.output[0]
else:
dq_node[0].input[0] = producer.output[0]

# Track QuantizeLinear node indices for cleanup
q_indices.append(node_idx)
Expand Down
9 changes: 5 additions & 4 deletions modelopt/onnx/trt_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -416,9 +416,10 @@ def interpret_trt_plugins_precision_flag(
# Will add Q/DQ nodes in the requested I/O indices
inp_precision_quant = [i for i, p in enumerate(inp_precision) if p in ["int8", "fp8"]]
out_precision_quant = [i for i, p in enumerate(out_precision) if p in ["int8", "fp8"]]
custom_ops_to_quantize[op_type] = {
"inp": inp_precision_quant,
"out": out_precision_quant,
}
if inp_precision_quant or out_precision_quant:
custom_ops_to_quantize[op_type] = {
"inp": inp_precision_quant,
"out": out_precision_quant,
}

return custom_ops_to_cast, custom_ops_to_quantize
Loading