toggle_grad

flow_matching/solver/discrete_solver.py

@@ -17,6 +17,7 @@
 from flow_matching.path import MixtureDiscreteProbPath
 
 from flow_matching.solver.solver import Solver
+from flow_matching.solver.utils import toggle_grad
 from flow_matching.utils import categorical, ModelWrapper
 from .utils import get_nearest_times
 
@@ -82,7 +83,7 @@ def __init__(
 
         self.source_distribution_p = source_distribution_p
 
-    @torch.no_grad()
+    @toggle_grad
     def sample(
         self,
         x_init: Tensor,

flow_matching/solver/ode_solver.py

@@ -11,6 +11,7 @@
 from torchdiffeq import odeint
 
 from flow_matching.solver.solver import Solver
+from flow_matching.solver.utils import toggle_grad
 from flow_matching.utils import gradient, ModelWrapper
 
 
@@ -27,7 +28,7 @@ def __init__(self, velocity_model: Union[ModelWrapper, Callable]):
         super().__init__()
         self.velocity_model = velocity_model
 
-    @torch.no_grad()
+    @toggle_grad
     def sample(
         self,
         x_init: Tensor,
@@ -101,7 +102,7 @@ def ode_func(t, x):
         else:
             return sol[-1]
 
-    @torch.no_grad()
+    @toggle_grad
     def compute_likelihood(
         self,
         x_1: Tensor,
@@ -174,16 +175,15 @@ def dynamics_func(t, states):
         y_init = (x_1, torch.zeros(x_1.shape[0], device=x_1.device))
         ode_opts = {"step_size": step_size} if step_size is not None else {}
 
-        with torch.no_grad():
-            sol, log_det = odeint(
-                dynamics_func,
-                y_init,
-                time_grid,
-                method=method,
-                options=ode_opts,
-                atol=atol,
-                rtol=rtol,
-            )
+        sol, log_det = odeint(
+            dynamics_func,
+            y_init,
+            time_grid,
+            method=method,
+            options=ode_opts,
+            atol=atol,
+            rtol=rtol,
+        )
 
         x_source = sol[-1]
         source_log_p = log_p0(x_source)

flow_matching/solver/riemannian_ode_solver.py

@@ -12,6 +12,7 @@
 from tqdm import tqdm
 
 from flow_matching.solver.solver import Solver
+from flow_matching.solver.utils import toggle_grad
 from flow_matching.utils import ModelWrapper
 from flow_matching.utils.manifolds import geodesic, Manifold
 
@@ -31,7 +32,7 @@ def __init__(self, manifold: Manifold, velocity_model: ModelWrapper):
         self.manifold = manifold
         self.velocity_model = velocity_model
 
-    @torch.no_grad()
+    @toggle_grad
     def sample(
         self,
         x_init: Tensor,

flow_matching/solver/utils.py

@@ -4,6 +4,8 @@
 # This source code is licensed under the CC-by-NC license found in the
 # LICENSE file in the root directory of this source tree.
 
+from functools import wraps
+
 import torch
 from torch import Tensor
 
@@ -17,3 +19,12 @@ def get_nearest_times(time_grid: Tensor, t_discretization: Tensor) -> Tensor:
     nearest_indices = distances.argmin(dim=1)
 
     return t_discretization[nearest_indices]
+
+
+def toggle_grad(func):
+    @wraps(func)
+    def wrapper(*args, enable_grad=False, **kwargs):
+        with torch.set_grad_enabled(enable_grad):
+            return func(*args, **kwargs)
+
+    return wrapper

tests/solver/test_ode_solver.py

@@ -23,9 +23,10 @@ def forward(self, x: torch.Tensor, t: torch.Tensor, **extras) -> torch.Tensor:
 class ConstantVelocityModel(ModelWrapper):
     def __init__(self):
         super().__init__(None)
+        self.a = torch.nn.Parameter(torch.tensor(1.0))
 
     def forward(self, x: torch.Tensor, t: torch.Tensor, **extras) -> torch.Tensor:
-        return x * 0.0 + 1.0
+        return x * 0.0 + self.a
 
 
 class TestODESolver(unittest.TestCase):
@@ -75,6 +76,27 @@ def test_sample_with_different_methods(self):
                     "The solution to the velocity field 3t^3 from 0 to 1 is incorrect.",
                 )
 
+    def test_gradients(self):
+        x_init = torch.tensor([1.0, 0.0])
+        step_size = 0.001
+        time_grid = torch.tensor([0.0, 1.0])
+
+        for method in ["euler", "dopri5", "midpoint", "heun3"]:
+            with self.subTest(method=method):
+                self.constant_velocity_model.zero_grad()
+                result = self.constant_velocity_solver.sample(
+                    x_init=x_init,
+                    step_size=step_size if method != "dopri5" else None,
+                    time_grid=time_grid,
+                    method=method,
+                    enable_grad=True,
+                )
+                loss = result.sum()
+                loss.backward()
+                self.assertAlmostEqual(
+                    self.constant_velocity_model.a.grad, 2.0, delta=1e-4
+                )
+
     def test_compute_likelihood(self):
         x_1 = torch.tensor([[0.0, 0.0]])
         step_size = 0.1
@@ -93,7 +115,7 @@ def dummy_log_p(x: Tensor) -> Tensor:
         self.assertEqual(x_1.shape[0], log_likelihood.shape[0])
 
     def test_compute_likelihood_exact_divergence(self):
-        x_1 = torch.tensor([[0.0, 0.0]])
+        x_1 = torch.tensor([[0.0, 0.0]], requires_grad=True)
         step_size = 0.1
 
         # Define a dummy log probability function
@@ -105,6 +127,7 @@ def dummy_log_p(x: Tensor) -> Tensor:
             log_p0=dummy_log_p,
             step_size=step_size,
             exact_divergence=True,
+            enable_grad=True,
         )
         self.assertIsInstance(log_likelihood, Tensor)
         self.assertEqual(x_1.shape[0], log_likelihood.shape[0])
@@ -114,6 +137,10 @@ def dummy_log_p(x: Tensor) -> Tensor:
         self.assertTrue(
             torch.allclose(x_1 - 1.0, x_0, atol=1e-2),
         )
+        log_likelihood.backward()
+        self.assertTrue(
+            torch.allclose(x_1.grad, torch.tensor([1.0, 1.0]), atol=1e-2),
+        )
 
 
 if __name__ == "__main__":