ONNX import: Remove extra flatten from Linalg_gemm

ONNX Gemm spec expects the input to be 2D. Flatten is not required
while importing Gemm, it will only lead to redundant flatten.

python/mxnet/contrib/onnx/onnx2mx/_op_translations.py

@@ -398,8 +398,7 @@ def linalg_gemm(attrs, inputs, proto_obj):
         alpha = attrs['alpha']
     if 'beta' in attrs:
         beta = attrs['beta']
-    flatten_a = symbol.flatten(inputs[0])
-    matmul_op = symbol.linalg_gemm2(A=flatten_a, B=inputs[1],
+    matmul_op = symbol.linalg_gemm2(A=inputs[0], B=inputs[1],
                                     transpose_a=trans_a, transpose_b=trans_b,
                                     alpha=alpha)
     gemm_op = symbol.broadcast_add(matmul_op, beta*inputs[2])