Fix GQA shape inference (#18723)

The shape inference is always returning before getting the chance to
infer the key/value outputs.
This commit is contained in:
Patrice Vignola 2023-12-10 21:36:19 -08:00 committed by GitHub
parent de32baeeef
commit 8d641229e6
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23

View file

@ -259,7 +259,6 @@ void GroupQueryAttentionTypeAndShapeInference(ONNX_NAMESPACE::InferenceContext&
*output_shape.add_dim() = query_dims[1];
*output_shape.add_dim() = query_dims[2];
updateOutputShape(ctx, 0, output_shape);
return;
} else {
fail_shape_inference("Missing input 2 (value)");
}