name ‘selective_scan_fn‘ is not defined运行出现这个错误

1.第一步先在这个代码中找到try以及except这部分，

2.在这个部分最下边编写一个打印输出

添加打印之后重新运行出现还是当前的错误

3.那么选择在第二部分pass之后另起一行复制下列代码，完成添加SelectiveScanFn这个类，就可以在此处调用了

class SelectiveScanFn(torch.autograd.Function):

    @staticmethod
    def forward(ctx, u, delta, A, B, C, D=None, z=None, delta_bias=None, delta_softplus=False,
                return_last_state=False):
        if u.stride(-1) != 1:
            u = u.contiguous()
        if delta.stride(-1) != 1:
            delta = delta.contiguous()
        if D is not None:
            D = D.contiguous()
        if B.stride(-1) != 1:
            B = B.contiguous()
        if C.stride(-1) != 1:
            C = C.contiguous()
        if z is not None and z.stride(-1) != 1:
            z = z.contiguous()
        if B.dim() == 3:
            B = rearrange(B, "b dstate l -> b 1 dstate l")
            ctx.squeeze_B = True
        if C.dim() == 3:
            C = rearrange(C, "b dstate l -> b 1 dstate l")
            ctx.squeeze_C = True
        out, x, *rest = selective_scan_cuda.fwd(u, delta, A, B, C, D, z, delta_bias, delta_softplus)
        ctx.delta_softplus = delta_softplus
        ctx.has_z = z is not None
        last_state = x[:, :, -1, 1::2]  # (batch, dim, dstate)
        if not ctx.has_z:
            ctx.save_for_backward(u, delta, A, B, C, D, delta_bias, x)
            return out if not return_last_state else (out, last_state)
        else:
            ctx.save_for_backward(u, delta, A, B, C, D, z, delta_bias, x, out)
            out_z = rest[0]
            return out_z if not return_last_state else (out_z, last_state)

    @staticmethod
    def backward(ctx, dout, *args):
        if not ctx.has_z:
            u, delta, A, B, C, D, delta_bias, x = ctx.saved_tensors
            z = None
            out = None
        else:
            u, delta, A, B, C, D, z, delta_bias, x, out = ctx.saved_tensors
        if dout.stride(-1) != 1:
            dout = dout.contiguous()
        # The kernel supports passing in a pre-allocated dz (e.g., in case we want to fuse the
        # backward of selective_scan_cuda with the backward of chunk).
        # Here we just pass in None and dz will be allocated in the C++ code.
        du, ddelta, dA, dB, dC, dD, ddelta_bias, *rest = selective_scan_cuda.bwd(
            u, delta, A, B, C, D, z, delta_bias, dout, x, out, None, ctx.delta_softplus,
            False  # option to recompute out_z, not used here
        )
        dz = rest[0] if ctx.has_z else None
        dB = dB.squeeze(1) if getattr(ctx, "squeeze_B", False) else dB
        dC = dC.squeeze(1) if getattr(ctx, "squeeze_C", False) else dC
        return (du, ddelta, dA, dB, dC,
                dD if D is not None else None,
                dz,
                ddelta_bias if delta_bias is not None else None,
                None,
                None)

def selective_scan_fn(u, delta, A, B, C, D=None, z=None, delta_bias=None, delta_softplus=False,
                      return_last_state=False):
    """if return_last_state is True, returns (out, last_state)
    last_state has shape (batch, dim, dstate). Note that the gradient of the last state is
    not considered in the backward pass.
    """
    return selective_scan_ref(u, delta, A, B, C, D, z, delta_bias, delta_softplus, return_last_state)

def selective_scan_ref(u, delta, A, B, C, D=None, z=None, delta_bias=None, delta_softplus=False,
                      return_last_state=False):
    """
    u: r(B D L)
    delta: r(B D L)
    A: c(D N) or r(D N)
    B: c(D N) or r(B N L) or r(B N 2L) or r(B G N L) or (B G N L)
    C: c(D N) or r(B N L) or r(B N 2L) or r(B G N L) or (B G N L)
    D: r(D)
    z: r(B D L)
    delta_bias: r(D), fp32

    out: r(B D L)
    last_state (optional): r(B D dstate) or c(B D dstate)
    """
    dtype_in = u.dtype
    u = u.float()
    delta = delta.float()
    if delta_bias is not None:
        delta = delta + delta_bias[..., None].float()
    if delta_softplus:
        delta = F.softplus(delta)
    batch, dim, dstate = u.shape[0], A.shape[0], A.shape[1]
    is_variable_B = B.dim() >= 3
    is_variable_C = C.dim() >= 3
    if A.is_complex():
        if is_variable_B:
            B = torch.view_as_complex(rearrange(B.float(), "... (L two) -> ... L two", two=2))
        if is_variable_C:
            C = torch.view_as_complex(rearrange(C.float(), "... (L two) -> ... L two", two=2))
    else:
        B = B.float()
        C = C.float()
    x = A.new_zeros((batch, dim, dstate))
    ys = []
    deltaA = torch.exp(torch.einsum('bdl,dn->bdln', delta, A))
    if not is_variable_B:
        deltaB_u = torch.einsum('bdl,dn,bdl->bdln', delta, B, u)
    else:
        if B.dim() == 3:
            deltaB_u = torch.einsum('bdl,bnl,bdl->bdln', delta, B, u)
        else:
            B = repeat(B, "B G N L -> B (G H) N L", H=dim // B.shape[1])
            deltaB_u = torch.einsum('bdl,bdnl,bdl->bdln', delta, B, u)
    if is_variable_C and C.dim() == 4:
        C = repeat(C, "B G N L -> B (G H) N L", H=dim // C.shape[1])
    last_state = None
    for i in range(u.shape[2]):
        x = deltaA[:, :, i] * x + deltaB_u[:, :, i]
        if not is_variable_C:
            y = torch.einsum('bdn,dn->bd', x, C)
        else:
            if C.dim() == 3:
                y = torch.einsum('bdn,bn->bd', x, C[:, :, i])
            else:
                y = torch.einsum('bdn,bdn->bd', x, C[:, :, :, i])
        if i == u.shape[2] - 1:
            last_state = x
        if y.is_complex():
            y = y.real * 2
        ys.append(y)
    y = torch.stack(ys, dim=2) # (batch dim L)
    out = y if D is None else y + u * rearrange(D, "d -> d 1")
    if z is not None:
        out = out * F.silu(z)
    out = out.to(dtype=dtype_in)
    return out if not return_last_state else (out, last_state)

def mamba_inner_fn(
        xz, conv1d_weight, conv1d_bias, x_proj_weight, delta_proj_weight,
        out_proj_weight, out_proj_bias,
        A, B=None, C=None, D=None, delta_bias=None, B_proj_bias=None,
        C_proj_bias=None, delta_softplus=True
):
    return mamba_inner_ref(xz, conv1d_weight, conv1d_bias, x_proj_weight, delta_proj_weight,
                           out_proj_weight, out_proj_bias,
                           A, B, C, D, delta_bias, B_proj_bias, C_proj_bias, delta_softplus)
 

最后重新运行这段代码，就可以运行成功了