-
Notifications
You must be signed in to change notification settings - Fork 2.4k
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.
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
[GPU] Implement bias on internal FC op #23317
Conversation
build_jenkins |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
overall, LGTM
@@ -40,6 +40,7 @@ ConvertFullyConnectedToFullyConnectedCompressed::ConvertFullyConnectedToFullyCon | |||
return in_ps.rank().is_static() && out_ps.rank().is_static() && in_ps.size() == 3 && out_ps.size() == 2; | |||
}; | |||
|
|||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
del
const ov::Output<Node> &decompression_scale, | ||
const ov::Output<Node> &decompression_zero_point, | ||
const ov::element::Type output_type = ov::element::undefined); | ||
|
||
FullyConnectedCompressed(const ov::Output<Node> &A, | ||
const ov::Output<Node> &B, | ||
const ov::Output<Node> &bias, | ||
const ov::Output<Node> &decompression_scale, | ||
const ov::element::Type output_type = ov::element::undefined); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Since Placeholder
op was added recently, we may now have single c-tor for this op with all arguments. Could you modify that too? Can be done in a separate PR if you want
@@ -44,7 +44,8 @@ MoveFCReshapeToWeights::MoveFCReshapeToWeights() { | |||
auto weights_input_m = std::make_shared<ov::pass::pattern::op::Or>(ov::OutputVector{reshape_m, transpose_m}); | |||
|
|||
auto data_m = any_input(); | |||
auto fully_connected_m = wrap_type<op::FullyConnected>({data_m, weights_input_m}); | |||
auto bias_m = any_input(); | |||
auto fully_connected_m = wrap_type<op::FullyConnected>({data_m, weights_input_m, bias_m}); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
nit: I think there's no big sense to have named variable for bias in such cases. Something like this should be shorter and won't lose any meaningful info
auto fully_connected_m = wrap_type<op::FullyConnected>({data_m, weights_input_m, any_input()});
Added bias semantics support for internal FC op
ebde6c8
Added bias semantics support for internal FC op ### Details: - Added bias semantics support for internal FC op
Added bias semantics support for internal FC op ### Details: - Added bias semantics support for internal FC op
Added bias semantics support for internal FC op ### Details: - Added bias semantics support for internal FC op
Added bias semantics support for internal FC op
Details: