benchmarks/transformer_fusion_patterns/benchmark.py [22:40]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        input = torch.randn(
            batch_size,
            seq_len,
            hidden_size,
            requires_grad=True,
            device=device,
            dtype=dtype,
        )
        bias = torch.randn(hidden_size, requires_grad=True, device=device, dtype=dtype)
        residual = torch.randn(
            batch_size,
            seq_len,
            hidden_size,
            requires_grad=False,
            device=device,
            dtype=dtype,
        )
        args = (input, bias, residual)
        return args
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



benchmarks/transformer_fusion_patterns/benchmark.py [88:106]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        input = torch.randn(
            batch_size,
            seq_len,
            hidden_size,
            requires_grad=True,
            device=device,
            dtype=dtype,
        )
        bias = torch.randn(hidden_size, requires_grad=True, device=device, dtype=dtype)
        residual = torch.randn(
            batch_size,
            seq_len,
            hidden_size,
            requires_grad=False,
            device=device,
            dtype=dtype,
        )
        args = (input, bias, residual)
        return args
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -