docs(mge/imperative): add docstring for GradManager

GitOrigin-RevId: 4c326206b83f7fb40f43ba3487bc76316440e1f8

imperative/python/megengine/autodiff/grad_manager.py

@@ -14,6 +14,51 @@ def get_backwarding_grad_manager():
 
 
 class GradManager:
+    r"""GradManager manages auto differentiation and all resources required to perform it.
+
+    Our auto differentiation framework requires that the user explicitly indicates when
+    the forward operations start and when all resources should be released. A typical usage of
+    GradManager is as follows:
+
+        .. codeblock::
+
+            gm = GradManager()
+            gm.attach(model.parameters())
+            with gm:
+                # forward operations
+                ...
+                # backward gradients
+                gm.backward(loss)
+
+    You can also use `record()` and `release()` method instead of `with` context:
+
+        .. codeblock::
+
+            gm = GradManager()
+            gm.attach(model.parameters())
+
+            gm.record()
+
+            # forward operations
+            ...
+            # backward gradients
+            gm.backward(loss)
+
+            gm.release()
+
+    Typically, in data parallel, we would like to average the gradients across
+    processes. Users will finally get the averaged gradients if an "AllReduce"
+    callback is registered as follows:
+
+        .. codeblock::
+
+            import megengine.distributed as dist
+
+            gm = GradManager()
+            gm.attach(model.parameters(), callback=dist.make_allreduce_cb("MEAN"))
+
+    """
+
     def __init__(self):
         self._call_back_dict = defaultdict(list)
         self._param_dict = dict()
@@ -23,6 +68,18 @@ class GradManager:
         self._gradients = dict()
 
     def attach(self, params, callbacks=None):
+        r"""Registers parameters that gradients should be calculated with respect to.
+        Callback Functions should have a signature like this:
+
+            .. codeblock::
+
+                def cb(param: Tensor, grad: Tensor) -> Tensor:
+                    # do something
+                    return grad
+
+        :param params: registered parameters
+        :param callbacks: list of callback functions
+        """
         if callbacks is None:
             callbacks = []
         if isinstance(callbacks, Callable):
@@ -38,6 +95,11 @@ class GradManager:
         return self
 
     def backward(self, ys, dys=None):
+        r"""Performs back-propagation and computes gradients.
+
+        :param ys: outputs of forward operators, e.g., the loss tensor
+        :param dys: derivatives of ys
+        """
         global backwarding_grad_manager
         cache = backwarding_grad_manager
         backwarding_grad_manager = self
@@ -71,6 +133,8 @@ class GradManager:
             backwarding_grad_manager = cache
 
     def record(self):
+        r"""Starts recording forward operations.
+        """
         if self._recording:
             raise RuntimeError("already recording")
         grad = Grad()
@@ -90,6 +154,8 @@ class GradManager:
         grad.__enter__()
 
     def release(self):
+        r"""Stops recording and releases resources for gradients calculation.
+        """
         if not self._recording:
             raise RuntimeError("not recording")
         self._stop_record()