Skip to content

[ET-VK][ez] Apply quantize op replacement to all argument nodes #15746

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
Nov 11, 2025
Merged

[ET-VK][ez] Apply quantize op replacement to all argument nodes #15746

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
33 changes: 16 additions & 17 deletions backends/vulkan/_passes/replace_qdq.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,24 +32,23 @@ def call(self, graph_module: torch.fx.GraphModule):
exir_ops.edge.et_vk.conv2d_q8ta_q8csw_q8to_dw.default,
exir_ops.edge.et_vk.add_q8ta_q8ta_q8to.default,
]:
# Replace quantize op feeding into conv2d (first argument is the quantized input)
quantized_input_node = node.args[0]
if isinstance(
quantized_input_node, torch.fx.Node
) and utils.is_quant_node(quantized_input_node):
# Get the arguments from the original quantize node
input_tensor = quantized_input_node.args[0]
scale = quantized_input_node.args[1]
zero_point = quantized_input_node.args[2]
for quantized_input_node in node.args:
if isinstance(
quantized_input_node, torch.fx.Node
) and utils.is_quant_node(quantized_input_node):
# Get the arguments from the original quantize node
input_tensor = quantized_input_node.args[0]
scale = quantized_input_node.args[1]
zero_point = quantized_input_node.args[2]

nodes_to_replace.append(
{
"old_node": quantized_input_node,
"new_target": exir_ops.edge.et_vk.quantize_q8ta_for_conv2d.default,
"args": (input_tensor, scale, zero_point),
"node_type": "quantize_input",
}
)
nodes_to_replace.append(
{
"old_node": quantized_input_node,
"new_target": exir_ops.edge.et_vk.quantize_q8ta_for_conv2d.default,
"args": (input_tensor, scale, zero_point),
"node_type": "quantize_input",
}
)

# Find dequantize ops that consume the output of this conv2d
for user in node.users:
Expand Down
Loading