[checkpointio] fix zero optimizer async save memory (#6151)

* [checkpointio] fix zero optimizer async save memory

* [checkpointio] fit new tensornvme api

* [checkpointio] fit new tensornvme api

colossalai/booster/plugin/low_level_zero_plugin.py

@@ -128,22 +128,20 @@ class LowLevelZeroCheckpointIO(TorchDDPCheckpointIO):
         # the `state_dict` in LowLevelZeroOptimizer has communication
         # if only the master rank collect state_dict and save,
         # the communication on each rank would not match
-        if use_async:
+        if use_async and self.coordinator.is_master():
             if id(optimizer) not in self.pinned_state_dicts:
                 self.pinned_state_dicts[id(optimizer)] = {}
             pinned_state_dicts = self.pinned_state_dicts[id(optimizer)]
         else:
             pinned_state_dicts = None
-        state_dict = optimizer.state_dict(pinned_state_dicts)
+        state_dict = optimizer.state_dict(pinned_state_dicts, only_on_master=True)
         if self.coordinator.is_master():
             if use_async:
                 from tensornvme.async_file_io import AsyncFileWriter
 
                 from colossalai.utils.safetensors import save_nested
 
-                f_writer = AsyncFileWriter(
+                f_writer = AsyncFileWriter(checkpoint, n_entries=self.N_WRITE_ENTRIES, backend="pthread")
-                    fp=open(checkpoint, "wb", buffering=0), n_entries=self.N_WRITE_ENTRIES, backend="pthread"
-                )
                 save_nested(f_writer, state_dict)
                 self.async_writers.append(f_writer)
             else:
@@ -192,13 +190,15 @@ class LowLevelZeroCheckpointIO(TorchDDPCheckpointIO):
         # state_dict only provide only 'param_groups'
         state_dict = optimizer.optim.state_dict()
         # state shard would be handled by the low-level zero optimizer
-        if use_async:
+        if use_async and self.coordinator.is_master():
             if id(optimizer) not in self.pinned_state_dicts:
                 self.pinned_state_dicts[id(optimizer)] = {}
             pinned_state_dicts = self.pinned_state_dicts[id(optimizer)]
         else:
             pinned_state_dicts = None
-        sharded_state = optimizer.state_dict_shard(max_shard_size=size_per_shard, pinned_state_dicts=pinned_state_dicts)
+        sharded_state = optimizer.state_dict_shard(
+            max_shard_size=size_per_shard, pinned_state_dicts=pinned_state_dicts, only_on_master=True
+        )
 
         # Preparing file paths and index file.
         states_name, save_index_file, param_group_file = get_optimizer_base_filenames(prefix, use_safetensors=use_async)
@@ -227,7 +227,7 @@ class LowLevelZeroCheckpointIO(TorchDDPCheckpointIO):
                     from colossalai.utils.safetensors import save_nested
 
                     f_writer = AsyncFileWriter(
-                        fp=open(checkpoint_file_path, "wb", buffering=0),
+                        checkpoint_file_path,
                         n_entries=self.N_WRITE_ENTRIES,
                         backend="pthread",
                     )

colossalai/checkpoint_io/checkpoint_io_base.py

@@ -72,7 +72,6 @@ class CheckpointIO(ABC):
     def _sync_io(self):
         for writer in self.async_writers:
             writer.synchronize()
-            writer.fp.close()
         self.async_writers.clear()
 
     def _sync_d2h(self):

colossalai/checkpoint_io/general_checkpoint_io.py

@@ -56,7 +56,7 @@ class GeneralCheckpointIO(CheckpointIO):
         if use_async:
             from tensornvme.async_file_io import AsyncFileWriter
 
-            writer = AsyncFileWriter(open(checkpoint, "wb", buffering=0), self.N_WRITE_ENTRIES, backend="pthread")
+            writer = AsyncFileWriter(checkpoint, self.N_WRITE_ENTRIES, backend="pthread")
             if id(model) not in self.pinned_state_dicts:
                 self.pinned_state_dicts[id(model)] = create_pinned_state_dict(state_dict)
             self.async_writers.append(writer)

colossalai/checkpoint_io/hybrid_parallel_checkpoint_io.py

@@ -690,9 +690,7 @@ class HybridParallelCheckpointIO(GeneralCheckpointIO):
 
                     from colossalai.utils.safetensors import move_and_save
 
-                    writer = AsyncFileWriter(
+                    writer = AsyncFileWriter(checkpoint, self.N_WRITE_ENTRIES, backend="pthread")
-                        open(checkpoint, "wb", buffering=0), self.N_WRITE_ENTRIES, backend="pthread"
-                    )
                     if id(model) not in self.pinned_state_dicts:
                         self.pinned_state_dicts[id(model)] = create_pinned_state_dict(state_dict)
                     self.async_writers.append(writer)

colossalai/checkpoint_io/utils.py

@@ -311,7 +311,7 @@ def async_save_state_dict_shards(
             index_file.append_weight_map(key, shard_file)
         checkpoint_file_path = os.path.join(checkpoint, shard_file)
 
-        writer = AsyncFileWriter(open(checkpoint_file_path, "wb", buffering=0), n_write_entries, backend="pthread")
+        writer = AsyncFileWriter(checkpoint_file_path, n_write_entries, backend="pthread")
         writers.append(writer)
 
         if pinned_state_dict is not None:

colossalai/zero/low_level/low_level_optim.py

@@ -776,7 +776,9 @@ class LowLevelZeroOptimizer(OptimizerWrapper):
 
         return {"state": packed_state, "param_groups": param_groups}
 
-    def state_dict(self, pinned_state_dicts: Optional[Dict[str, Dict[str, torch.Tensor]]] = None) -> Dict:
+    def state_dict(
+        self, pinned_state_dicts: Optional[Dict[str, Dict[str, torch.Tensor]]] = None, only_on_master: bool = False
+    ) -> Dict:
         """Return a state_dict same with DDP
 
         Returns:
@@ -785,16 +787,22 @@ class LowLevelZeroOptimizer(OptimizerWrapper):
         zero_state = dict()
         device = get_accelerator().get_current_device()
         for param, state in self.optim.state.items():
-            if pinned_state_dicts is not None and param not in pinned_state_dicts:
-                pinned_state_dicts[param] = {}
-            zero_state[param] = copy.deepcopy(state)
-            for k, v in state.items():
-                if isinstance(v, torch.Tensor) and k != "step":
             working_param = self.master_to_working_param[id(param)]
             pg = self.param_to_pg[working_param]
+            if not only_on_master or get_nd_rank(pg) == 0:
+                zero_state[param] = copy.deepcopy(state)
+            else:
+                zero_state[param] = {}
+
+            if pinned_state_dicts is not None and param not in pinned_state_dicts:
+                pinned_state_dicts[param] = {}
+
+            for k, v in state.items():
+                if isinstance(v, torch.Tensor) and k != "step":
                     gathered_tensor = torch.empty(v.numel() * get_nd_world_size(pg), device=device, dtype=v.dtype)
                     all_gather_into_flat_tensor_nd(gathered_tensor, v.to(device).flatten(), pg)
                     param_state = gathered_tensor[: working_param.numel()].reshape_as(working_param)
+                    if not only_on_master or get_nd_rank(pg) == 0:
                         if pinned_state_dicts is not None and k not in pinned_state_dicts[param]:
                             pinned_state_dicts[param][k] = torch.empty_like(param_state, pin_memory=True, device="cpu")
                         if pinned_state_dicts is not None:
@@ -837,7 +845,10 @@ class LowLevelZeroOptimizer(OptimizerWrapper):
         self.optim.load_state_dict(zero_state_dict)
 
     def state_dict_shard(
-        self, max_shard_size: int = 1024, pinned_state_dicts: Optional[Dict[str, Dict[str, torch.Tensor]]] = None
+        self,
+        max_shard_size: int = 1024,
+        pinned_state_dicts: Optional[Dict[str, Dict[str, torch.Tensor]]] = None,
+        only_on_master: bool = False,
     ) -> Iterator[Tuple[Dict, int]]:
         """Returns dictionaries containing a whole state of the module one by one. The max size of dictionary shard is specified by ``max_shard_size``.
            Only include the 'state' in state_dict.
@@ -862,20 +873,26 @@ class LowLevelZeroOptimizer(OptimizerWrapper):
                 cnt += 1
         for param_idx, states in local_states.items():
             current_block_size = 0
-            current_block = copy.deepcopy(states)
             if pinned_state_dicts is not None and param_idx not in pinned_state_dicts:
                 pinned_state_dicts[param_idx] = {}
             master_param = idx2master[param_idx]
             working_param = self.master_to_working_param[id(master_param)]
             pg = self.param_to_pg[working_param]
+            if not only_on_master or get_nd_rank(pg) == 0:
+                current_block = copy.deepcopy(states)
+            else:
+                current_block = {}
 
             for k, v in states.items():
                 if isinstance(v, torch.Tensor) and k != "step":
                     state_tensor = torch.empty(v.numel() * get_nd_world_size(pg), device=device, dtype=v.dtype)
                     all_gather_into_flat_tensor_nd(state_tensor, v.to(device).flatten(), pg)
                     state_tensor = state_tensor[: working_param.numel()].reshape_as(working_param)
+                    if not only_on_master or get_nd_rank(pg) == 0:
                         if pinned_state_dicts is not None and k not in pinned_state_dicts[param_idx]:
-                        pinned_state_dicts[param_idx][k] = torch.empty_like(state_tensor, pin_memory=True, device="cpu")
+                            pinned_state_dicts[param_idx][k] = torch.empty_like(
+                                state_tensor, pin_memory=True, device="cpu"
+                            )
                         if pinned_state_dicts is not None:
                             pinned_state_dicts[param_idx][k].copy_(state_tensor)
                             current_block[k] = pinned_state_dicts[param_idx][k]

tests/test_checkpoint_io/test_safetensors_async_io.py

@@ -10,6 +10,7 @@ try:
 except ModuleNotFoundError:
     raise ModuleNotFoundError("Please install tensornvme to use NVMeOptimizer")
 
+
 from colossalai.testing import check_state_dict_equal
 from colossalai.utils import get_current_device
 
@@ -110,20 +111,20 @@ def test_save_load():
         }
 
         optimizer_saved_path = f"{tempdir}/save_optimizer.safetensors"
-        f_writer = AsyncFileWriter(fp=open(optimizer_saved_path, "wb"), n_entries=191, backend="pthread")
+        f_writer = AsyncFileWriter(optimizer_saved_path, n_entries=191, backend="pthread")
         save_nested(f_writer, optimizer_state_dict)
         f_writer.sync_before_step()
         f_writer.synchronize()
-        f_writer.fp.close()
+        del f_writer
         load_state_dict = load_flat(optimizer_saved_path)
         check_state_dict_equal(load_state_dict, optimizer_state_dict)
 
         optimizer_shard_saved_path = f"{tempdir}/save_optimizer_shard.safetensors"
-        f_writer = AsyncFileWriter(fp=open(optimizer_shard_saved_path, "wb"), n_entries=191, backend="pthread")
+        f_writer = AsyncFileWriter(optimizer_shard_saved_path, n_entries=191, backend="pthread")
         save_nested(f_writer, optimizer_state_dict["state"])
         f_writer.sync_before_step()
         f_writer.synchronize()
-        f_writer.fp.close()
+        del f_writer
         load_state_dict_shard = load_flat(optimizer_shard_saved_path)
         check_state_dict_equal(load_state_dict_shard, optimizer_state_dict["state"])
 
@@ -133,21 +134,21 @@ def test_save_load():
             "module.weight2": torch.rand((1024, 1024)),
         }
         model_saved_path = f"{tempdir}/save_model.safetensors"
-        f_writer = AsyncFileWriter(fp=open(model_saved_path, "wb"), n_entries=191, backend="pthread")
+        f_writer = AsyncFileWriter(model_saved_path, n_entries=191, backend="pthread")
         save(f_writer, model_state_dict)
         f_writer.sync_before_step()
         f_writer.synchronize()
-        f_writer.fp.close()
+        del f_writer
         load_state_dict = load_file(model_saved_path)
         check_state_dict_equal(model_state_dict, load_state_dict)
 
         model_state_dict_cuda = {k: v.to(get_current_device()) for k, v in model_state_dict.items()}
         model_state_pinned = {k: v.pin_memory() for k, v in model_state_dict.items()}
         model_saved_path = f"{tempdir}/save_model_cuda.safetensors"
-        f_writer = AsyncFileWriter(fp=open(model_saved_path, "wb"), n_entries=191, backend="pthread")
+        f_writer = AsyncFileWriter(model_saved_path, n_entries=191, backend="pthread")
         move_and_save(f_writer, model_state_dict_cuda, model_state_pinned)
         f_writer.sync_before_step()
         f_writer.synchronize()
-        f_writer.fp.close()
+        del f_writer
         load_state_dict = load_file(model_saved_path)
         check_state_dict_equal(model_state_dict, load_state_dict)