Operator: aten._log_softmax.default
cnt: 1, ((T([256, 256008], f16), 1, False), {})
Operator: aten._log_softmax_backward_data.default
cnt: 1, ((T([256, 256008], f16), T([256, 256008], f16), 1, f16), {})
Operator: aten._softmax.default
cnt: 24, ((T([32, 128, 128], f16), -1, False), {})
Operator: aten._softmax_backward_data.default
cnt: 24, ((T([32, 128, 128], f16), T([32, 128, 128], f16), -1, f16), {})
Operator: aten._to_copy.default
cnt: 1, ((T([128, 128], f32),), {'dtype': f16})
cnt: 1, ((T([2, 1, 128, 128], f16, stride=(0, 16384, 128, 1)),), {'dtype': f16, 'layout': torch.strided, 'device': 'cuda'})
cnt: 1, ((T([2, 128], b8),), {'dtype': i32})
cnt: 1, ((T([2, 128], i64),), {'dtype': i32, 'layout': torch.strided, 'device': 'cuda'})
cnt: 1, ((T([2, 128], i32),), {'dtype': i64})
Operator: aten._unsafe_view.default
cnt: 72, ((T([2, 128, 16, 64], f16), [2, 128, 1024]), {})
cnt: 1, ((T([256, 256008], f16), [2, 128, 256008]), {})
cnt: 24, ((T([2, 16, 128, 64], f16), [32, 128, 64]), {})
cnt: 24, ((T([2, 128, 1024], f16), [256, 1024]), {})
Operator: aten.add.Tensor
cnt: 1, ((T([128], i64), 1), {})
cnt: 1, ((T([2, 128], i32), 0), {})
cnt: 1, ((T([2, 128], i64), 1), {})
cnt: 145, ((T([2, 128, 1024], f16), T([2, 128, 1024], f16)), {})
cnt: 24, ((T([2, 16, 128, 128], f16), T([2, 1, 128, 128], f16)), {})
cnt: 1, ((T([256008, 1024], f16), T([256008, 1024], f16)), {})
Operator: aten.addmm.default
cnt: 96, ((T([1024], f16), T([256, 1024], f16), T([1024, 1024], f16, stride=(1, 1024))), {})
cnt: 24, ((T([4096], f16), T([256, 1024], f16), T([1024, 4096], f16, stride=(1, 1024))), {})
cnt: 24, ((T([1024], f16), T([256, 4096], f16), T([4096, 1024], f16, stride=(1, 4096))), {})
Operator: aten.bmm.default
cnt: 48, ((T([32, 128, 64], f16), T([32, 64, 128], f16, stride=(8192, 1, 64))), {})
cnt: 48, ((T([32, 128, 128], f16), T([32, 128, 64], f16)), {})
cnt: 24, ((T([32, 128, 128], f16, stride=(16384, 1, 128)), T([32, 128, 64], f16)), {})
cnt: 24, ((T([32, 64, 128], f16, stride=(8192, 1, 64)), T([32, 128, 128], f16)), {})
Operator: aten.clone.default
cnt: 2, ((T([2, 128], i64),), {})
cnt: 1, ((T([2, 127], i64, stride=(128, 1)),), {})
Operator: aten.copy_.default
cnt: 2, ((T([2, 128], i64), T([2, 128], i64)), {})
cnt: 1, ((T([2, 127], i64, stride=(128, 1)), T([2, 127], i64)), {})
Operator: aten.cumsum.default
cnt: 1, ((T([2, 128], i32), 1), {})
Operator: aten.embedding.default
cnt: 1, ((T([256008, 1024], f16), T([2, 128], i64), 1), {})
Operator: aten.embedding_dense_backward.default
cnt: 1, ((T([2, 128, 1024], f16), T([2, 128], i64), 256008, 1, False), {})
Operator: aten.fill_.Tensor
cnt: 1, ((T([2], i64, stride=(128,)), T([], i64)), {})
Operator: aten.gelu.default
cnt: 24, ((T([2, 128, 4096], f16),), {})
Operator: aten.gelu_backward.default
cnt: 24, ((T([2, 128, 4096], f16), T([2, 128, 4096], f16)), {})
Operator: aten.index_select.default
cnt: 1, ((T([2050, 1024], f16), 0, T([256], i64)), {})
Operator: aten.lt.Tensor
cnt: 1, ((T([128], i64), T([128, 1], i64)), {})
Operator: aten.masked_fill_.Scalar
cnt: 1, ((T([128, 128], f32), T([128, 128], b8), 0), {})
Operator: aten.mm.default
cnt: 1, ((T([256, 1024], f16), T([1024, 256008], f16, stride=(1, 1024))), {})
cnt: 1, ((T([256008, 256], f16, stride=(1, 256008)), T([256, 1024], f16)), {})
cnt: 1, ((T([256, 256008], f16), T([256008, 1024], f16)), {})
cnt: 24, ((T([256, 1024], f16), T([1024, 4096], f16)), {})
cnt: 24, ((T([1024, 256], f16, stride=(1, 1024)), T([256, 4096], f16)), {})
cnt: 24, ((T([256, 4096], f16), T([4096, 1024], f16)), {})
cnt: 24, ((T([4096, 256], f16, stride=(1, 4096)), T([256, 1024], f16)), {})
cnt: 96, ((T([256, 1024], f16), T([1024, 1024], f16)), {})
cnt: 96, ((T([1024, 256], f16, stride=(1, 1024)), T([256, 1024], f16)), {})
Operator: aten.mul.Tensor
cnt: 2, ((T([2, 128, 1024], f16), 32.0), {})
cnt: 1, ((T([2, 128], i32), T([2, 128], i32)), {})
cnt: 48, ((T([2, 128, 1024], f16), 0.125), {})
Operator: aten.native_layer_norm.default
cnt: 49, ((T([2, 128, 1024], f16), [1024], T([1024], f16), T([1024], f16), 1e-05), {})
Operator: aten.native_layer_norm_backward.default
cnt: 49, ((T([2, 128, 1024], f16), T([2, 128, 1024], f16), [1024], T([2, 128, 1], f32), T([2, 128, 1], f32), T([1024], f16), T([1024], f16), [True, True, True]), {})
Operator: aten.ne.Scalar
cnt: 1, ((T([2, 128], i64), 1), {})
Operator: aten.new_zeros.default
cnt: 1, ((T([2, 128], i64), [2, 128]), {'dtype': i64, 'layout': torch.strided, 'device': 'cuda', 'pin_memory': False})
Operator: aten.nll_loss_backward.default
cnt: 1, ((T([], f16), T([256, 256008], f16), T([256], i64), None, 1, -100, T([], f16)), {})
Operator: aten.nll_loss_forward.default
cnt: 1, ((T([256, 256008], f16), T([256], i64), None, 1, -100), {})
Operator: aten.sum.SymInt
cnt: 120, ((T([256, 1024], f16), [0], True), {})
cnt: 24, ((T([256, 4096], f16), [0], True), {})
