Add softplus torch backend

cirkit/backend/torch/parameters/nodes.py

@@ -728,6 +728,17 @@ def forward(self, x: Tensor) -> Tensor:
         return torch.clamp(x, min=self.vmin, max=self.vmax)
 
 
+class TorchSoftplusParameter(TorchEntrywiseParameterOp):
+    """Softmax reparameterization.
+
+    Range: (0, + inf), 0 available if input is masked.
+    Constraints: Positive.
+    """
+
+    def forward(self, x: Tensor) -> Tensor:
+        return torch.nn.functional.softplus(x)
+
+
 class TorchConjugateParameter(TorchEntrywiseParameterOp):
     """Conjugate parameterization."""
 

cirkit/backend/torch/rules/parameters.py

@@ -31,6 +31,7 @@
     TorchScaledSigmoidParameter,
     TorchSigmoidParameter,
     TorchSoftmaxParameter,
+    TorchSoftplusParameter,
     TorchSquareParameter,
     TorchSumParameter,
     TorchTensorParameter,
@@ -61,6 +62,7 @@
     ScaledSigmoidParameter,
     SigmoidParameter,
     SoftmaxParameter,
+    SoftplusParameter,
     SquareParameter,
     SumParameter,
     TensorParameter,
@@ -188,6 +190,13 @@ def compile_clamp_parameter(compiler: "TorchCompiler", p: ClampParameter) -> Tor
     return TorchClampParameter(in_shape, vmin=p.vmin, vmax=p.vmax)
 
 
+def compile_softplus_parameter(
+    compiler: "TorchCompiler", p: SoftplusParameter
+) -> TorchSoftplusParameter:
+    (in_shape,) = p.in_shapes
+    return TorchSoftplusParameter(in_shape)
+
+
 def compile_conjugate_parameter(
     compiler: "TorchCompiler", p: ClampParameter
 ) -> TorchConjugateParameter:
@@ -285,6 +294,7 @@ def compile_polynomial_differential(
     SigmoidParameter: compile_sigmoid_parameter,
     ScaledSigmoidParameter: compile_scaled_sigmoid_parameter,
     ClampParameter: compile_clamp_parameter,
+    SoftplusParameter: compile_softplus_parameter,
     ConjugateParameter: compile_conjugate_parameter,
     ReduceSumParameter: compile_reduce_sum_parameter,
     ReduceProductParameter: compile_reduce_product_parameter,

cirkit/templates/utils.py

@@ -25,6 +25,7 @@
     ParameterFactory,
     SigmoidParameter,
     SoftmaxParameter,
+    SoftplusParameter,
     TensorParameter,
     UnaryParameterOp,
 )
@@ -189,6 +190,8 @@ def name_to_parameter_activation(
             if "vmin" not in kwargs:
                 kwargs["vmin"] = 1e-18
             return functools.partial(ClampParameter, **kwargs)
+        case "softplus":
+            return functools.partial(SoftplusParameter, **kwargs)
         case _:
             raise ValueError