[zero] global model data memory tracer (#360)

colossalai/utils/commons/singleton_meta.py

@@ -0,0 +1,18 @@
+class SingletonMeta(type):
+    """
+    The Singleton class can be implemented in different ways in Python. Some
+    possible methods include: base class, decorator, metaclass. We will use the
+    metaclass because it is best suited for this purpose.
+    """
+
+    _instances = {}
+
+    def __call__(cls, *args, **kwargs):
+        """
+        Possible changes to the value of the `__init__` argument do not affect
+        the returned instance.
+        """
+        if cls not in cls._instances:
+            instance = super().__call__(*args, **kwargs)
+            cls._instances[cls] = instance
+        return cls._instances[cls]

colossalai/utils/memory_tracer/allocator.py

@@ -0,0 +1,60 @@
+import torch
+from colossalai.utils.commons.singleton_meta import SingletonMeta
+from colossalai.zero.sharded_param import ShardedTensor
+
+from typing import Union
+
+
+def col_tensor_mem_usage(t: Union[torch.Tensor, ShardedTensor]) -> int:
+    if isinstance(t, ShardedTensor):
+        target = t.payload
+    else:
+        target = t
+    return target.numel() * target.element_size()
+
+
+class ModelDataTracer(metaclass=SingletonMeta):
+    """
+    A singleton to trace model data usage during runtime.
+    """
+
+    def __init__(self) -> None:
+        self._cpu_usage = 0
+        self._cuda_usage = 0
+
+    def trace_tensor(self, t: torch.Tensor):
+        mem_use = col_tensor_mem_usage(t)
+        if t.device.type == 'cpu':
+            self._cpu_usage += mem_use
+        elif t.device.type == 'cuda':
+            self._cuda_usage += mem_use
+        else:
+            raise RuntimeError
+
+    def detach_tensor(self, t: torch.Tensor):
+        mem_use = col_tensor_mem_usage(t)
+        if t.device.type == 'cpu':
+            self._cpu_usage -= mem_use
+        elif t.device.type == 'cuda':
+            self._cuda_usage -= mem_use
+        else:
+            raise RuntimeError
+
+    @property
+    def cpu_usage(self):
+        return self._cpu_usage
+
+    @property
+    def cuda_usage(self):
+        return self._cuda_usage
+
+
+GLOBAL_MODEL_DATA_TRACER = ModelDataTracer()
+
+
+def col_allocate_payload(device: torch.device) -> torch.Tensor:
+    pass
+
+
+def col_release_payload(t: torch.Tensor):
+    pass

colossalai/zero/init_ctx/init_context.py

@@ -4,6 +4,7 @@ import torch
 from colossalai.utils.cuda import get_current_device
 from colossalai.zero.shard_utils import BaseShardStrategy
 from colossalai.zero.sharded_param import ShardedParamV2
+from colossalai.utils.memory_tracer.allocator import GLOBAL_MODEL_DATA_TRACER
 
 
 # Inserts _post_init_method at the end of init method
@@ -76,11 +77,16 @@ class InsertPostInitMethodToModuleSubClasses(object):
 
 
 class ZeroInitContext(InsertPostInitMethodToModuleSubClasses):
-    """
+    r"""
     A context to initialize model.
     1. Convert the model to fp16.
     2. The paramaters of the module are adapted to type ShardedParameter.
     3. Shard the param and grad according to flags.
+    rm_torch_payload_on_the_fly:
+    True: remove tensor payload on param.data after module init finished.
+    False: remove tensor payload on param.data afther the context exist.
+            This is used when you add some logic to operate tensors in __init__ of module.
+            See torchvision resnet18.
     """
 
     def __init__(self,
@@ -134,5 +140,7 @@ class ZeroInitContext(InsertPostInitMethodToModuleSubClasses):
 
             if self.shard_param:
                 self.shard_strategy.shard(tensor_list=[param.col_attr._data_sharded_tensor])
+                GLOBAL_MODEL_DATA_TRACER.trace_tensor(param.col_attr._data_sharded_tensor.payload)
             if param.col_attr.grad and self.shard_grad:
                 self.shard_strategy.shard(tensor_list=[param.col_attr._grad_sharded_tensor])
+                GLOBAL_MODEL_DATA_TRACER.trace_tensor(param.col_attr._grad_sharded_tensor.payload)

colossalai/zero/sharded_param/sharded_tensor.py

@@ -7,7 +7,7 @@ class ShardedTensor(object):
 
     def __init__(self, tensor: torch.Tensor, process_group: Optional[dist.ProcessGroup] = None) -> None:
         r"""
-        A tensor sharded in multiple processes.
+        A tensor sharded in multiple processes. Constructed from an existing torch.Tensor instance.
         """
         self._payload = tensor
         self.process_group = process_group

tests/test_zero_data_parallel/test_init_context.py

@@ -13,7 +13,8 @@ from colossalai.zero.shard_utils.tensor_shard_strategy import \
     TensorShardStrategy
 from tests.components_to_test.registry import non_distributed_component_funcs
 
-from common import CONFIG, Net
+from common import CONFIG
+from colossalai.utils.memory_tracer.allocator import GLOBAL_MODEL_DATA_TRACER
 
 
 def run_dist(rank, world_size, port):
@@ -33,9 +34,12 @@ def run_dist(rank, world_size, port):
             assert param.col_attr.data.is_sharded
             assert param.col_attr.data.payload.device.type == 'cuda'
 
+    print(f'cuda usgae {GLOBAL_MODEL_DATA_TRACER.cuda_usage}')
+    assert (GLOBAL_MODEL_DATA_TRACER.cuda_usage > 0)
+
 
 @pytest.mark.dist
-@pytest.mark.parametrize("world_size", [1, 2, 4])
+@pytest.mark.parametrize("world_size", [1, 4])
 def test_zero_init_context(world_size):
     run_func = partial(run_dist, world_size=world_size, port=free_port())
     mp.spawn(run_func, nprocs=world_size)