[fx] add more meta_registry for MetaTensor execution. (#2000)

* [sc] add examples for auto checkpoint.

* merge upstream

* [fx] add more meta_registry for MetaTensor execution.

colossalai/fx/_meta_registrations.py

@@ -3,7 +3,7 @@
 # refer to https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/native_functions.yaml
 # for more meta_registrations
 
-from typing import List, Optional, Tuple, Union
+from typing import Callable, List, Optional, Tuple, Union
 
 import torch
 from torch.utils._pytree import tree_map
@@ -179,6 +179,42 @@ def meta_adaptive_avg_pool2d_backward(
     return grad_input
 
 
+# ================================ RNN =============================================
+# https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/cudnn/RNN.cpp
+@register_meta(aten._cudnn_rnn.default)
+def meta_cuda_rnn(
+    input: torch.Tensor,
+    weight: torch.Tensor,
+    weight_stride0: int,
+    weight_buf: torch.Tensor,
+    hx: torch.Tensor,
+    cx: Optional[torch.Tensor] = None,
+    *args,
+    **kwargs,
+):
+    if cx is not None:
+        return torch.empty_like(input), torch.empty_like(hx), torch.empty_like(cx)
+    else:
+        return torch.empty_like(input), torch.empty_like(hx), torch.empty((), device='meta')
+
+
+# https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/cudnn/RNN.cpp
+@register_meta(aten._cudnn_rnn_backward.default)
+def meta_cudnn_rnn_backward(input: torch.Tensor,
+                            weight: torch.Tensor,
+                            weight_stride0: int,
+                            hx: torch.Tensor,
+                            cx: Optional[torch.Tensor] = None,
+                            *args,
+                            **kwargs):
+    print(input, weight, hx, cx)
+    grad_input = torch.empty_like(input)
+    grad_weight = torch.empty_like(weight)
+    grad_hx = torch.empty_like(hx)
+    grad_cx = torch.empty_like(cx) if cx is not None else torch.empty((), device='meta')
+    return grad_input, grad_weight, grad_hx, grad_cx
+
+
 # https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/Activation.cpp
 # ============================== Activations =======================================
 @register_meta(aten.relu.default)
@@ -186,6 +222,11 @@ def meta_relu(input: torch.Tensor):
     return torch.empty_like(input)
 
 
+@register_meta(aten.prelu.default)
+def meta_prelu(input: torch.Tensor, weight: torch.Tensor):
+    return torch.empty_like(input)
+
+
 @register_meta(aten.hardswish.default)
 def meta_hardswish(input: torch.Tensor):
     return torch.empty_like(input)
@@ -278,12 +319,18 @@ def meta_ln_backward(dY: torch.Tensor, input: torch.Tensor, normalized_shape, me
 
 
 # ================================== Misc ==========================================
-#https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/native_functions.yaml
+# https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/native_functions.yaml
 @register_meta(aten.roll.default)
 def meta_roll(input: torch.Tensor, shifts, dims):
     return input
 
 
+# https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/Scalar.cpp
+@register_meta(aten._local_scalar_dense.default)
+def meta_local_scalar_dense(self: torch.Tensor):
+    return 0
+
+
 # https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/TensorCompare.cpp
 @register_meta(aten.where.self)
 def meta_where_self(condition: torch.Tensor, self: torch.Tensor, other: torch.Tensor):
@@ -317,7 +364,7 @@ def meta_index_Tensor(self, indices):
     indices = result
     assert len(indices) <= self.ndim, f"too many indices for tensor of dimension {self.ndim} (got {len(indices)})"
     # expand_outplace
-    import torch._refs as refs    # avoid import cycle in mypy
+    import torch._refs as refs
 
     indices = list(refs._maybe_broadcast(*indices))
     # add missing null tensors

colossalai/fx/profiler/tensor.py

@@ -128,3 +128,13 @@ class MetaTensor(torch.Tensor):
         if device is not None:
             result = MetaTensor(result, fake_device=device)
         return result
+
+    def cpu(self, *args, **kwargs):
+        if self.device.type == 'cpu':
+            return self.to(*args, **kwargs)
+        return self.to(*args, device='cpu', **kwargs)
+
+    def cuda(self, *args, **kwargs):
+        if self.device.type == 'cuda':
+            return self.to(*args, **kwargs)
+        return self.to(*args, device='cuda', **kwargs)

colossalai/fx/tracer/_symbolic_trace.py

@@ -20,28 +20,25 @@ def symbolic_trace(
     Given an ``nn.Module`` or function instance ``root``, this function will return a ``ColoGraphModule``
     constructed by recording operations seen while tracing through ``root``.
 
-    With ``meta_args`` and ``concrete_args``, we can trace the model that are untraceable subject to control flow.
-    If specified using ``meta_args`` only, the tracing can be done ahead of time.
+    With ``meta_args``, we can trace the model that are untraceable subject to control flow. If specified using
+    ``meta_args`` only, the tracing can be done ahead of time.
 
-    Note that both ``meta_args`` and ``concrete_args`` are kwargs, which contains the key of the argument's names
-    and the value of the argument's values.
+    Note that ``meta_args`` are kwargs, which contains the key of the argument's names and the value of the
+    argument's values.
 
     Uses:
         >>> model = ...
 
         # if this works
-        >>> gm = symbolic_trace(model)
+        >>> gm = symbolic_trace(model, concrete_args=concrete_args)
 
         # else try this
-        >>> gm = symbolic_trace(model, meta_args={'x': torch.rand(1, 3, 224, 224, device='meta')})
-
-        # else try this
-        >>> gm = symbolic_trace(model, concrete_args={'x': torch.rand(1, 3, 224, 224)})
+        >>> gm = symbolic_trace(model, concrete_args=concrete_args, meta_args={'x': torch.rand(1, 3, 224, 224, device='meta')})
 
     Args:
         root (Union[torch.nn.Module, Callable[..., Any]]): Module or function to be traced and converted
             into a Graph representation.
-        concrete_args (Optional[Dict[str, Any]], optional): Inputs to be partially specialized. Defaults to None.
+        concrete_args (Optional[Dict[str, Any]], optional): Concrete arguments to be used for tracing.
         meta_args (Optional[Dict[str, Any]], optional): Inputs to be partially specialized, special for ``ColoTracer``.
             Defaults to None.
 
@@ -52,7 +49,6 @@ def symbolic_trace(
         This API is still under development and can incur some bugs. Feel free to report any bugs to the Colossal-AI team.
 
     """
-    tracer = ColoTracer()
-    graph = tracer.trace(root, concrete_args, meta_args)
-    name = (root.__class__.__name__ if isinstance(root, torch.nn.Module) else root.__name__)
-    return ColoGraphModule(tracer.root, graph, name)
+    graph = ColoTracer().trace(root, concrete_args=concrete_args, meta_args=meta_args)
+    name = root.__class__.__name__ if isinstance(root, torch.nn.Module) else root.__name__
+    return ColoGraphModule(root, graph, name)

examples/tutorial/auto_parallel/bench_utils.py

@@ -18,13 +18,11 @@ def bench(gm: torch.fx.GraphModule,
           data_gen: Callable,
           num_steps: int = 5) -> Tuple[int, int]:
     """Benchmarking a given graph module
-
     Args:
         gm (torch.fx.GraphModule): The graph module to benchmark.
         criterion (torch.nn.Module): Loss function.
         data_gen (Callable): Data generator.
         num_steps (int, optional): Number of test steps. Defaults to 5.
-
     Returns:
         Tuple[int, int]: peak memory in MB and step time in MS.
     """
@@ -69,7 +67,6 @@ def bench_rotor(gm: torch.fx.GraphModule,
                 start_factor: int = 4) -> Tuple[np.array, list, list]:
     """Auto Checkpoint Rotor Algorithm benchmarking
     Benchmarks the Auto Checkpoint Rotor Algorithm for a given graph module and data.
-
     Args:
         gm (torch.fx.GraphModule): The graph module to benchmark.
         criterion (torch.nn.Module): Loss function.
@@ -79,7 +76,6 @@ def bench_rotor(gm: torch.fx.GraphModule,
         free_memory (int, optional): Max memory budget in Byte. Defaults to torch.cuda.mem_get_info()[0].
         start_factor (int, optional): Start memory budget factor for benchmark, the start memory budget
         will be free_memory / start_factor. Defaults to 4.
-
     Returns:
         Tuple[np.array, list, list]: return budgets vector (MB), peak memory vector (MB), step time vector (MS).
     """