deploy: SimplexLab/TorchJD@e890e65

Author: ValerianRey

latest/_sources/examples/amp.rst.txt

@@ -12,7 +12,7 @@ case, the losses) should preferably be scaled with a `GradScaler
 following example shows the resulting code for a multi-task learning use-case.
 
 .. code-block:: python
-    :emphasize-lines: 2, 17, 27, 34, 36-38
+    :emphasize-lines: 2, 17, 27, 34-37
 
     import torch
     from torch.amp import GradScaler
@@ -48,10 +48,10 @@ following example shows the resulting code for a multi-task learning use-case.
             loss2 = loss_fn(output2, target2)
 
         scaled_losses = scaler.scale([loss1, loss2])
-        optimizer.zero_grad()
         mtl_backward(losses=scaled_losses, features=features, aggregator=aggregator)
         scaler.step(optimizer)
         scaler.update()
+        optimizer.zero_grad()
 
 .. hint::
     Within the ``torch.autocast`` context, some operations may be done in ``float16`` type. For

latest/_sources/examples/basic_usage.rst.txt

@@ -59,12 +59,6 @@ We can now compute the losses associated to each element of the batch.
 
 The last steps are similar to gradient descent-based optimization, but using the two losses.
 
-Reset the ``.grad`` field of each model parameter:
-
-.. code-block:: python
-
-    optimizer.zero_grad()
-
 Perform the Jacobian descent backward pass:
 
 .. code-block:: python
@@ -81,3 +75,9 @@ Update each parameter based on its ``.grad`` field, using the ``optimizer``:
     optimizer.step()
 
 The model's parameters have been updated!
+
+As usual, you should now reset the ``.grad`` field of each model parameter:
+
+.. code-block:: python
+
+    optimizer.zero_grad()

latest/_sources/examples/iwmtl.rst.txt

@@ -10,7 +10,7 @@ this Gramian to reweight the gradients and resolve conflict entirely.
 The following example shows how to do that.
 
 .. code-block:: python
-    :emphasize-lines: 5-6, 18-20, 31-32, 34-35, 37-38, 41-42
+    :emphasize-lines: 5-6, 18-20, 31-32, 34-35, 37-38, 40-41
 
     import torch
     from torch.nn import Linear, MSELoss, ReLU, Sequential
@@ -51,10 +51,10 @@ The following example shows how to do that.
         # Obtain the weights that lead to no conflict between reweighted gradients
         weights = weighting(gramian)  # shape: [16, 2]
 
-        optimizer.zero_grad()
         # Do the standard backward pass, but weighted using the obtained weights
         losses.backward(weights)
         optimizer.step()
+        optimizer.zero_grad()
 
 .. note::
     In this example, the tensor of losses is a matrix rather than a vector. The gramian is thus a

latest/_sources/examples/iwrm.rst.txt

@@ -64,19 +64,19 @@ batch of data. When minimizing per-instance losses (IWRM), we use either autojac
             for x, y in zip(X, Y):
                 y_hat = model(x).squeeze(dim=1)  # shape: [16]
                 loss = loss_fn(y_hat, y)  # shape: [] (scalar)
-                optimizer.zero_grad()
                 loss.backward()
 
 
                 optimizer.step()
+                optimizer.zero_grad()
 
         In this baseline example, the update may negatively affect the loss of some elements of the
         batch.
 
     .. tab-item:: autojac
 
         .. code-block:: python
-            :emphasize-lines: 5-6, 12, 16, 21, 23
+            :emphasize-lines: 5-6, 12, 16, 21-22
 
             import torch
             from torch.nn import Linear, MSELoss, ReLU, Sequential
@@ -99,19 +99,19 @@ batch of data. When minimizing per-instance losses (IWRM), we use either autojac
             for x, y in zip(X, Y):
                 y_hat = model(x).squeeze(dim=1)  # shape: [16]
                 losses = loss_fn(y_hat, y)  # shape: [16]
-                optimizer.zero_grad()
                 backward(losses, aggregator)
 
 
                 optimizer.step()
+                optimizer.zero_grad()
 
         Here, we compute the Jacobian of the per-sample losses with respect to the model parameters
         and use it to update the model such that no loss from the batch is (locally) increased.
 
     .. tab-item:: autogram (recommended)
 
         .. code-block:: python
-            :emphasize-lines: 5-6, 12, 16-17, 21, 23-25
+            :emphasize-lines: 5-6, 12, 16-17, 21-24
 
             import torch
             from torch.nn import Linear, MSELoss, ReLU, Sequential
@@ -134,11 +134,11 @@ batch of data. When minimizing per-instance losses (IWRM), we use either autojac
             for x, y in zip(X, Y):
                 y_hat = model(x).squeeze(dim=1)  # shape: [16]
                 losses = loss_fn(y_hat, y)  # shape: [16]
-                optimizer.zero_grad()
                 gramian = engine.compute_gramian(losses)  # shape: [16, 16]
                 weights = weighting(gramian)  # shape: [16]
                 losses.backward(weights)
                 optimizer.step()
+                optimizer.zero_grad()
 
         Here, the per-sample gradients are never fully stored in memory, leading to large
         improvements in memory usage and speed compared to autojac, in most practical cases. The

latest/_sources/examples/lightning_integration.rst.txt

@@ -11,7 +11,7 @@ The following code example demonstrates a basic multi-task learning setup using
 <../docs/autojac/mtl_backward>` at each training iteration.
 
 .. code-block:: python
-    :emphasize-lines: 9-10, 18, 32
+    :emphasize-lines: 9-10, 18, 31
 
     import torch
     from lightning import LightningModule, Trainer
@@ -43,9 +43,9 @@ The following code example demonstrates a basic multi-task learning setup using
             loss2 = mse_loss(output2, target2)
 
             opt = self.optimizers()
-            opt.zero_grad()
             mtl_backward(losses=[loss1, loss2], features=features, aggregator=UPGrad())
             opt.step()
+            opt.zero_grad()
 
         def configure_optimizers(self) -> OptimizerLRScheduler:
             optimizer = Adam(self.parameters(), lr=1e-3)

latest/_sources/examples/monitoring.rst.txt

@@ -63,6 +63,6 @@ they have a negative inner product).
         loss1 = loss_fn(output1, target1)
         loss2 = loss_fn(output2, target2)
 
-        optimizer.zero_grad()
         mtl_backward(losses=[loss1, loss2], features=features, aggregator=aggregator)
         optimizer.step()
+        optimizer.zero_grad()

latest/_sources/examples/mtl.rst.txt

@@ -19,7 +19,7 @@ vectors of dimension 10, and their corresponding scalar labels for both tasks.
 
 
 .. code-block:: python
-    :emphasize-lines: 5-6, 19, 33
+    :emphasize-lines: 5-6, 19, 32
 
     import torch
     from torch.nn import Linear, MSELoss, ReLU, Sequential
@@ -52,9 +52,9 @@ vectors of dimension 10, and their corresponding scalar labels for both tasks.
         loss1 = loss_fn(output1, target1)
         loss2 = loss_fn(output2, target2)
 
-        optimizer.zero_grad()
         mtl_backward(losses=[loss1, loss2], features=features, aggregator=aggregator)
         optimizer.step()
+        optimizer.zero_grad()
 
 .. note::
     In this example, the Jacobian is only with respect to the shared parameters. The task-specific

latest/_sources/examples/partial_jd.rst.txt

@@ -41,8 +41,8 @@ first ``Linear`` layer, thereby reducing memory usage and computation time.
     for x, y in zip(X, Y):
         y_hat = model(x).squeeze(dim=1)  # shape: [16]
         losses = loss_fn(y_hat, y)  # shape: [16]
-        optimizer.zero_grad()
         gramian = engine.compute_gramian(losses)
         weights = weighting(gramian)
         losses.backward(weights)
         optimizer.step()
+        optimizer.zero_grad()

latest/_sources/examples/rnn.rst.txt

@@ -6,7 +6,7 @@ element of the output sequences. If the gradients of these losses are likely to
 descent can be leveraged to enhance optimization.
 
 .. code-block:: python
-    :emphasize-lines: 5-6, 10, 17, 20
+    :emphasize-lines: 5-6, 10, 17, 19
 
     import torch
     from torch.nn import RNN
@@ -26,9 +26,9 @@ descent can be leveraged to enhance optimization.
         output, _ = rnn(input)  # output is of shape [5, 3, 20].
         losses = ((output - target) ** 2).mean(dim=[1, 2])  # 1 loss per sequence element.
 
-        optimizer.zero_grad()
         backward(losses, aggregator, parallel_chunk_size=1)
         optimizer.step()
+        optimizer.zero_grad()
 
 .. note::
     At the time of writing, there seems to be an incompatibility between ``torch.vmap`` and

latest/docs/autogram/engine/index.html

@@ -349,12 +349,12 @@ <h1>Engine<a class="headerlink" href="#engine" title="Link to this heading">¶</
     <span class="n">output</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span><span class="o">.</span><span class="n">squeeze</span><span class="p">(</span><span class="n">dim</span><span class="o">=</span><span class="mi">1</span><span class="p">)</span>  <span class="c1"># shape: [16]</span>
     <span class="n">losses</span> <span class="o">=</span> <span class="n">criterion</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>  <span class="c1"># shape: [16]</span>
 
-    <span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
-<span class="hll">    <span class="n">gramian</span> <span class="o">=</span> <span class="n">engine</span><span class="o">.</span><span class="n">compute_gramian</span><span class="p">(</span><span class="n">losses</span><span class="p">)</span>  <span class="c1"># shape: [16, 16]</span>
-</span><span class="hll">    <span class="n">weights</span> <span class="o">=</span> <span class="n">weighting</span><span class="p">(</span><span class="n">gramian</span><span class="p">)</span>  <span class="c1"># shape: [16]</span>
+    <span class="n">gramian</span> <span class="o">=</span> <span class="n">engine</span><span class="o">.</span><span class="n">compute_gramian</span><span class="p">(</span><span class="n">losses</span><span class="p">)</span>  <span class="c1"># shape: [16, 16]</span>
+<span class="hll">    <span class="n">weights</span> <span class="o">=</span> <span class="n">weighting</span><span class="p">(</span><span class="n">gramian</span><span class="p">)</span>  <span class="c1"># shape: [16]</span>
 </span><span class="hll">    <span class="n">losses</span><span class="o">.</span><span class="n">backward</span><span class="p">(</span><span class="n">weights</span><span class="p">)</span>
-</span>    <span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
-</pre></div>
+</span><span class="hll">    <span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
+</span><span class="hll">    <span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
+</span></pre></div>
 </div>
 <p>This is equivalent to just calling <code class="docutils literal notranslate"><span class="pre">torchjd.autojac.backward(losses,</span> <span class="pre">UPGrad())</span></code>. However,
 since the Jacobian never has to be entirely in memory, it is often much more