[hotfix] ZeroDDP use new process group (#1333)

* process group supports getting ranks in group

* chunk mgr receives a process group

* update unit test

* fix unit tests

colossalai/gemini/chunk.py

@@ -4,9 +4,8 @@ from dataclasses import dataclass
 from enum import Enum
 from typing import Optional, Dict, List
 
-from colossalai.core import global_context as gpc
-from colossalai.context import ParallelMode
 from colossalai.utils import get_current_device
+from colossalai.tensor import ProcessGroup as ColoProcessGroup
 
 
 class TensorState(Enum):
@@ -65,14 +64,16 @@ class Chunk:
     def __init__(self,
                  chunk_size: int,
                  src_rank: int,
+                 process_group: ColoProcessGroup,
                  dtype: torch.dtype,
                  init_device: Optional[torch.device] = None,
                  force_data_on_cuda: bool = False) -> None:
         self.size = chunk_size
         self.utilized_size = 0
         self.src_rank = src_rank
-        self.is_src_rank = gpc.get_local_rank(ParallelMode.DATA) == src_rank
-        self.global_src_rank = gpc.get_ranks_in_group(ParallelMode.DATA)[src_rank]
+        self.process_group = process_group
+        self.is_src_rank = process_group.dp_local_rank() == src_rank
+        self.global_src_rank = process_group.get_ranks_in_dp()[src_rank]
         self.dtype = dtype
         device = init_device or get_current_device()
         if force_data_on_cuda:
@@ -150,7 +151,7 @@ class Chunk:
         if not self.is_src_rank:
             alloc_storage(self._payload)
         self.move_device(get_current_device(), update_ptr=False)
-        dist.broadcast(self.data, self.global_src_rank, group=gpc.get_group(ParallelMode.DATA))
+        dist.broadcast(self.data, self.global_src_rank, group=self.process_group.dp_process_group())
 
         # update tensor meta info
         self._update_tensors_ptr()
@@ -193,9 +194,9 @@ class Chunk:
         """
         self.move_device(get_current_device(), update_ptr=False)
         if is_all_reduce:
-            dist.all_reduce(self.data, group=gpc.get_group(ParallelMode.DATA))
+            dist.all_reduce(self.data, group=self.process_group.dp_process_group())
         else:
-            dist.reduce(self.data, self.global_src_rank, group=gpc.get_group(ParallelMode.DATA))
+            dist.reduce(self.data, self.global_src_rank, group=self.process_group.dp_process_group())
         self._update_tensors_ptr()
         self._update_tensors_state(TensorState.HOLD)
 
@@ -216,7 +217,7 @@ class Chunk:
         # invalid calls will be ignored and nothing changes
         if (self.tensors_info[tensor].state, tensor_state) not in STATE_TRANS:
             # print(
-            #     f'WARNING: Rank{gpc.get_global_rank()} apply invalid state trans: {self.tensors_info[tensor].state} to {tensor_state}'
+            #     f'WARNING: Rank{self.process_group.rank()} apply invalid state trans: {self.tensors_info[tensor].state} to {tensor_state}'
             # )
             return
         self.tensors_info[tensor].state = tensor_state

colossalai/gemini/chunk_mgr.py

@@ -2,9 +2,8 @@ import torch
 from typing import Optional, Dict, Deque, Set, List, Tuple, Iterable
 from collections import deque
 
-from colossalai.context import ParallelMode
-from colossalai.core import global_context as gpc
 from colossalai.utils import get_current_device
+from colossalai.tensor import ProcessGroup as ColoProcessGroup
 from .chunk import Chunk, ChunkFullError, TensorState
 
 
@@ -20,10 +19,13 @@ class ChunkManager:
 
     def __init__(self,
                  chunk_size: Optional[int],
+                 process_group: ColoProcessGroup,
                  enable_distributed_storage: bool = False,
                  init_device: Optional[torch.device] = None) -> None:
         assert chunk_size is None or chunk_size > 0
+        assert isinstance(process_group, ColoProcessGroup)
         self.chunk_size = chunk_size
+        self.process_group = process_group
         self.enable_distributed_storage = enable_distributed_storage
         self.device = init_device or get_current_device()
         self.chunk_groups: Dict[str, Deque[Chunk]] = {}
@@ -69,6 +71,7 @@ class ChunkManager:
             src_rank = self._get_next_src_rank(group_name)
             chunk = Chunk(chunk_size,
                           src_rank,
+                          self.process_group,
                           tensor.dtype,
                           self.device,
                           force_data_on_cuda=self.groups_force_data_on_cuda[group_name])
@@ -89,17 +92,17 @@ class ChunkManager:
     def _get_next_src_rank(self, group_name: str) -> int:
         if not self.enable_distributed_storage:
             # the chunk is owned by the current rank if no distributed storage is enabled
-            return gpc.get_local_rank(ParallelMode.DATA)
+            return self.process_group.dp_local_rank()
         if self.chunk_size is None:
             if group_name not in self.rank_load:
-                self.rank_load[group_name] = torch.zeros(gpc.get_world_size(ParallelMode.DATA), dtype=torch.int64)
+                self.rank_load[group_name] = torch.zeros(self.process_group.dp_world_size(), dtype=torch.int64)
 
             # the process owning the tensor will be the process with the smallest number of elements
             src_rank = torch.argmin(self.rank_load[group_name]).item()
         else:
             # chunk is owned by processes in a round-robin fashion
             chunk_idx = len(self.chunk_groups[group_name])
-            src_rank = chunk_idx % gpc.get_world_size(ParallelMode.DATA)
+            src_rank = chunk_idx % self.process_group.dp_world_size()
         return src_rank
 
     def access_chunk(self, chunk: Chunk) -> None:
@@ -222,7 +225,7 @@ class ChunkManager:
         self.lazy_release_tensors.clear()
 
     def __repr__(self) -> str:
-        msg = f'Rank {gpc.get_local_rank(ParallelMode.DATA)}:\n'
+        msg = f'Rank {self.process_group.dp_local_rank()}:\n'
         msg += 'Total memory: ' + ', '.join([f'{k}={v}B' for k, v in self.total_mem.items()]) + '\n'
         for group_name, group in self.chunk_groups.items():
             msg += f'Group {group_name}:\n'

colossalai/nn/parallel/data_parallel.py

@@ -118,7 +118,7 @@ class ColoDDP(torch.nn.Module):
             return empty_grad
 
         else:
-            #TODO(jiaruifang) fixme
+            # TODO(jiaruifang) fixme
             self.process_group.set_cpu_groups()
             dist.all_reduce(grad, group=self.process_group.cpu_dp_process_group())
             return grad
@@ -191,11 +191,8 @@ class ZeroDDP(ColoDDP):
             For more details, see the API reference of ``GeminiManager``.
     """
 
-    def __init__(self,
-                 module: torch.nn.Module,
-                 gemini_manager: GeminiManager,
-                 process_group: Optional[ColoProcessGroup] = None) -> None:
-        super().__init__(module.half(), process_group=process_group)
+    def __init__(self, module: torch.nn.Module, gemini_manager: GeminiManager) -> None:
+        super().__init__(module.half(), process_group=gemini_manager.chunk_manager.process_group)
         self.gemini_manager = gemini_manager
         self.chunk_manager = gemini_manager.chunk_manager
         self.param_op_hook = ZeROHookV2(gemini_manager)

colossalai/tensor/process_group.py

@@ -171,3 +171,9 @@ class ProcessGroup:
     def cpu_tp_process_group(self):
         assert self._has_cpu_groups
         return PYTORCHPGDICT_.get(self._tp_rank_list, 'gloo')
+
+    def get_ranks_in_dp(self):
+        return self._dp_rank_list
+
+    def get_ranks_in_tp(self):
+        return self._tp_rank_list

tests/test_ddp/test_ddp_ignore_params.py

@@ -33,11 +33,11 @@ def init_ddp(module: torch.nn.Module) -> ColoDDP:
 
 
 def init_ddpv2(module: torch.nn.Module, use_chunk: bool = False) -> ZeroDDP:
+    pg = ProcessGroup()
     chunk_size = ChunkManager.search_chunk_size(module, 64, 2) if use_chunk else None
-    chunk_manager = ChunkManager(chunk_size)
+    chunk_manager = ChunkManager(chunk_size, pg)
     gemini_manager = GeminiManager('cuda', chunk_manager)
-    pg = ProcessGroup()
-    return ZeroDDP(module, gemini_manager, pg)
+    return ZeroDDP(module, gemini_manager)
 
 
 class Net(torch.nn.Module):

tests/test_ddp/test_ddp_state_dict.py

@@ -28,11 +28,11 @@ def init_ddp(module: torch.nn.Module) -> ColoDDP:
 
 
 def init_ddpv2(module: torch.nn.Module, use_chunk: bool = False, use_zero: bool = False) -> ZeroDDP:
+    pg = ProcessGroup()
     chunk_size = ChunkManager.search_chunk_size(module, 64, 4) if use_chunk else None
-    chunk_manager = ChunkManager(chunk_size, enable_distributed_storage=use_zero)
+    chunk_manager = ChunkManager(chunk_size, pg, enable_distributed_storage=use_zero)
     gemini_manager = GeminiManager('cuda', chunk_manager)
-    pg = ProcessGroup()
-    return ZeroDDP(module, gemini_manager, process_group=pg)
+    return ZeroDDP(module, gemini_manager)
 
 
 def run_state_dict(ddp_init_func: Callable[[torch.nn.Module], ColoDDP]):

tests/test_tensor/test_chunk.py

@@ -7,8 +7,7 @@ from functools import partial
 from colossalai.gemini import ChunkManager
 from colossalai.testing import rerun_if_address_is_in_use, parameterize
 from colossalai.utils import free_port
-from colossalai.core import global_context as gpc
-from colossalai.context import ParallelMode
+from colossalai.tensor import ProcessGroup as ColoProcessGroup
 
 
 def check_has_params(params: List[torch.Tensor], has_tensors: List[bool]):
@@ -38,12 +37,13 @@ TOTAL_MEM = {True: {True: [512, 512], False: [1024, 1024]}, False: {True: [512,
 @parameterize('use_chunk', [False, True])
 @parameterize('use_zero', [False, True])
 def run_chunk_zero(use_chunk, use_zero):
-    rank = gpc.get_local_rank(ParallelMode.DATA)
+    pg = ColoProcessGroup()
+    rank = pg.rank()
     if rank == 0:
         print(f'use_chunk={use_chunk}, use_zero={use_zero}')
     params = [torch.rand(8, 8) for _ in range(3)]
     chunk_size = 128 if use_chunk else None
-    chunk_manager = ChunkManager(chunk_size, enable_distributed_storage=use_zero)
+    chunk_manager = ChunkManager(chunk_size, pg, enable_distributed_storage=use_zero)
     chunk_manager.create_group('param')
     assert chunk_manager.total_mem['cpu'] == 0
     assert chunk_manager.total_mem['cuda'] == 0

tests/test_tensor/test_zero_optim.py

@@ -31,8 +31,6 @@ def check_param_equal(model, torch_model, pg: ProcessGroup):
 def check_grad_equal(model, torch_model, pg: ProcessGroup):
     for (n, p), (tn, tp) in zip(model.named_parameters(), torch_model.named_parameters()):
         if p.grad is not None:
-            torch.distributed.barrier()
-            print(torch.distributed.get_rank(), p.grad)
             assert tensor_shard_equal(tp.grad.to(dtype=p.grad.dtype, device=p.grad.device), p.grad,
                                       pg.tp_local_rank(), pg.tp_world_size()), \
                 f'{tp.grad} vs {p.grad}\n{n}:\n\t{tp.grad.shape} vs {p.grad.shape} in {pg.rank()}'
@@ -63,9 +61,9 @@ def init_1d_col_spec(model, pg: ProcessGroup):
             p.set_tensor_spec(*spec)
 
 
-@parameterize('use_chunk', [False])
-@parameterize('use_zero', [False])
-@parameterize('placement_policy', ['cuda'])
+@parameterize('use_chunk', [False, True])
+@parameterize('use_zero', [False, True])
+@parameterize('placement_policy', ['cuda', 'cpu'])
 def run_gpt(use_chunk, use_zero, placement_policy, tp_init_spec_func=None):
     set_seed(42)
     get_components_func = non_distributed_component_funcs.get_callable('gpt2')
@@ -92,10 +90,11 @@ def run_gpt(use_chunk, use_zero, placement_policy, tp_init_spec_func=None):
 
     chunk_size = ChunkManager.search_chunk_size(model, 8192, 8) if use_chunk else None
     chunk_manager = ChunkManager(chunk_size,
+                                 pg,
                                  enable_distributed_storage=use_zero,
                                  init_device=GeminiManager.get_default_device(placement_policy))
     gemini_manager = GeminiManager(placement_policy, chunk_manager)
-    model = ZeroDDP(model, gemini_manager, pg)
+    model = ZeroDDP(model, gemini_manager)
     optim = HybridAdam(model.parameters(), lr=1e-3)
     optim = ZeroOptimizer(optim, model, initial_scale=1)
 
@@ -104,7 +103,7 @@ def run_gpt(use_chunk, use_zero, placement_policy, tp_init_spec_func=None):
     torch_model, torch_optim = convert_to_apex_amp(torch_model, torch_optim, amp_config)
     torch_model = DDP(torch_model, device_ids=[pg.rank()], process_group=pg.dp_process_group())
 
-    # print(chunk_manager)
+    print(chunk_manager)
     check_param_equal(model, torch_model, pg)
 
     model.eval()
@@ -129,13 +128,12 @@ def run_dist(rank, world_size, port):
     colossalai.launch(config=config, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
     if world_size == 4:
         run_gpt(tp_init_spec_func=init_1d_col_spec)
-        # run_gpt(tp_init_spec_func=init_1d_row_spec)
+        run_gpt(tp_init_spec_func=init_1d_row_spec)
     else:
         run_gpt(tp_init_spec_func=init_1d_col_spec)
 
 
 @pytest.mark.dist
-@pytest.mark.skip("buggy test")
 @pytest.mark.parametrize('world_size', [1, 4])
 @rerun_if_address_is_in_use()
 def test_gpt(world_size):

tests/test_zero/test_zero_optim_state_dict.py

@@ -20,13 +20,14 @@ from colossalai.tensor import ProcessGroup
 
 
 def init_zero(model, use_chunk, use_zero, placement_policy):
+    pg = ProcessGroup()
     chunk_size = ChunkManager.search_chunk_size(model, 8192, 8) if use_chunk else None
     chunk_manager = ChunkManager(chunk_size,
+                                 pg,
                                  enable_distributed_storage=use_zero,
                                  init_device=GeminiManager.get_default_device(placement_policy))
     gemini_manager = GeminiManager(placement_policy, chunk_manager)
-    pg = ProcessGroup()
-    return ZeroDDP(model, gemini_manager, pg)
+    return ZeroDDP(model, gemini_manager)
 
 
 def run_step(model, optim, criterion, data, label):