[moe] implement submesh initialization

Author: botbw

colossalai/booster/plugin/moe_hybrid_parallel_plugin.py

@@ -1,6 +1,7 @@
 import warnings
 from types import MethodType
 from typing import Callable, Optional, OrderedDict, Tuple
+import numpy as np
 
 import torch
 import torch.distributed as dist
@@ -64,6 +65,14 @@ def __init__(
             overlap_communication = True
             warnings.warn(WARN_STR + " Please make sure of this.")
 
+        self.param_info = param_info
+        self.stage_manager = model.stage_manager
+        self.shared_params = model.shared_params
+        self.dp_pg = dp_process_group
+    
+        if use_pipeline:
+            reinitialize_optimizer(optimizer, model)
+
         pg_param_list = {
             dp_process_group: list(filter(lambda p: not is_moe_tensor(p), model.parameters())),
             moe_dp_group: list(filter(is_moe_tensor, model.parameters())),
@@ -116,17 +125,16 @@ def __init__(self, ep_size: int, moe_tp_size: int = 1, force_overlap_comm=False,
             raise NotImplementedError
 
         world_size = dist.get_world_size()
-
-        self.moe_dp_size = world_size // (ep_size * moe_tp_size)
+        self.moe_dp_size = world_size // (self.pp_size * ep_size * moe_tp_size)
         self.ep_size = ep_size
         self.moe_tp_size = moe_tp_size
 
-        self.moe_pg_mesh = ProcessGroupMesh(self.moe_dp_size, self.ep_size, self.moe_tp_size)
-        self.moe_dp_axis, self.ep_axis, self.moe_tp_axis = 0, 1, 2
+        if self.pp_size * self.moe_dp_size * self.ep_size * self.moe_tp_size != world_size:
+            raise ValueError(
+                f"world_size={world_size} is not divisible by pp_size={self.pp_size} * moe_dp_size={self.moe_dp_size} * ep_size={self.ep_size} * moe_tp_size={self.moe_tp_size}"
+            )
 
-        self.moe_dp_group = self.moe_pg_mesh.get_group_along_axis(self.moe_dp_axis)
-        self.ep_group = self.moe_pg_mesh.get_group_along_axis(self.ep_axis)
-        self.moe_tp_group = self.moe_pg_mesh.get_group_along_axis(self.moe_tp_axis)
+        self._init_moe_param_comm()
 
         self.logger.info(f"{type(self).__name__}: {self.ep_size=} {self.moe_dp_size=} {self.moe_tp_size=}", ranks=[0])
 
@@ -136,6 +144,52 @@ def __init__(self, ep_size: int, moe_tp_size: int = 1, force_overlap_comm=False,
 
         self.force_overlap_comm = force_overlap_comm
 
+    def _init_moe_param_comm(self):
+        self.moe_dp_group = None
+        self.ep_group = None
+        self.moe_tp_group = None
+        
+        # create submesh for ep, moe_dp, moe_tp
+        ranks_by_pp_stage = self.pg_mesh.get_group_along_axis(
+            [self.dp_axis, self.tp_axis, self.sp_axis], return_ranks_by_group=True
+        )
+
+        global_rank = self.pg_mesh.rank
+        pp_rank = self.pg_mesh.coordinate(self.pp_axis)
+
+        # create groups from submesh
+        for stage_idx, stage_rank in enumerate(ranks_by_pp_stage):
+            # axis 0 is dp, axis 1 is tp, axis 2 is sp
+            submesh = np.array(stage_rank).reshape(self.moe_dp_size, self.ep_size, self.moe_tp_size)
+
+            # hardcode here since we only have 3 axis
+            # moe_dp_group
+            for ep_idx in range(self.ep_size):
+                for moe_tp_idx in range(self.moe_tp_size):
+                    moe_dp_ranks = submesh[:, ep_idx, moe_tp_idx].flatten().tolist()
+                    group = dist.new_group(moe_dp_ranks)
+                    if pp_rank == stage_idx and global_rank in moe_dp_ranks:
+                        assert self.moe_dp_group is None
+                        self.moe_dp_group = group
+            # ep_group
+            for moe_dp_idx in range(self.moe_dp_size):
+                for moe_tp_idx in range(self.moe_tp_size):
+                    ep_ranks = submesh[moe_dp_idx, :, moe_tp_idx].flatten().tolist()
+                    group = dist.new_group(ep_ranks)
+                    if pp_rank == stage_idx and global_rank in ep_ranks:
+                        assert self.ep_group is None
+                        self.ep_group = group
+            # moe_tp_group
+            for moe_dp_idx in range(self.moe_dp_size):
+                for ep_idx in range(self.ep_size):
+                    moe_tp_ranks = submesh[moe_dp_idx, ep_idx, :].flatten().tolist()
+                    group = dist.new_group(moe_tp_ranks)
+                    if pp_rank == stage_idx and global_rank in moe_tp_ranks:
+                        assert self.moe_tp_group is None
+                        self.moe_tp_group = group
+
+        self.logger.info(f"rank {dist.get_rank()} moe_dp_group {dist.get_process_group_ranks(self.moe_dp_group)} ep_group {dist.get_process_group_ranks(self.ep_group)} moe_tp_group {dist.get_process_group_ranks(self.moe_tp_group)}")
+
     def get_checkpoint_io(self) -> MoECheckpointIO:
         return MoECheckpointIO(
             self.dp_group, self.pp_group, self.tp_group, self.ep_group, self.moe_dp_group, self.zero_stage

colossalai/cluster/process_group_mesh.py

@@ -209,13 +209,15 @@ def create_group_along_axis(
         axis: Union[int, List[int]],
         indices_at_axis: Optional[Union[List[int], List[List[int]]]] = None,
         backend: Optional[str] = None,
-    ) -> ProcessGroup:
+        return_ranks_by_group: bool = False
+    ) -> Union[ProcessGroup, List[Tuple[int, ...]]]:
         """Create all process groups along the given axis, and return the one which the current process belongs to.
 
         Args:
             axis (int): Axis along which the process groups are created.
             indices_at_axis (Optional[List[int]], optional): Indices at the axis. Defaults to None.
             backend (Optional[str], optional): Backend of the process group. Defaults to None.
+            return_ranks_by_group (bool): Whether to return all ranks by group for creating submesh. Defaults to False.
 
         Returns:
             ProcessGroup: The process group along the given axis which the current process belongs to.
@@ -235,25 +237,35 @@ def create_group_along_axis(
         # the choices on the axis are reduced to 1, since it's determined by `indices_at_axis`
         for ax in axis:
             reduced_shape[ax] = 1
-        target_group = None
-        # use Cartesian product to generate all combinations of coordinates
-        for base_coord in itertools.product(*[range(s) for s in reduced_shape]):
-            coords_in_group = ProcessGroupMesh.get_coords_along_axis(base_coord, axis, indices_at_axis)
-            ranks_in_group = tuple([ProcessGroupMesh.ravel(coord, self._shape) for coord in coords_in_group])
-            group = self._get_group(ranks_in_group, backend=backend)
-            if self._rank in ranks_in_group:
-                target_group = group
-        return target_group
+        if return_ranks_by_group:
+            ranks_by_group = []
+            # use Cartesian product to generate all combinations of coordinates
+            for base_coord in itertools.product(*[range(s) for s in reduced_shape]):
+                coords_in_group = ProcessGroupMesh.get_coords_along_axis(base_coord, axis, indices_at_axis)
+                ranks_in_group = tuple([ProcessGroupMesh.ravel(coord, self._shape) for coord in coords_in_group])
+                ranks_by_group.append(ranks_in_group)
+            return ranks_by_group
+        else:
+            target_group = None
+            # use Cartesian product to generate all combinations of coordinates
+            for base_coord in itertools.product(*[range(s) for s in reduced_shape]):
+                coords_in_group = ProcessGroupMesh.get_coords_along_axis(base_coord, axis, indices_at_axis)
+                ranks_in_group = tuple([ProcessGroupMesh.ravel(coord, self._shape) for coord in coords_in_group])
+                group = self._get_group(ranks_in_group, backend=backend)
+                if self._rank in ranks_in_group:
+                    target_group = group
+            return target_group
 
     def get_group_along_axis(
-        self, axis: Union[int, List[int]], indices_at_axis: Optional[List[int]] = None, backend: Optional[str] = None
-    ) -> ProcessGroup:
+        self, axis: Union[int, List[int]], indices_at_axis: Optional[List[int]] = None, backend: Optional[str] = None, return_ranks_by_group: bool = False
+    ) -> Union[ProcessGroup, List[Tuple[int, ...]]]:
         """Get the process group along the given axis which the current process belongs to. If the process group doesn't exist, it will be created.
 
         Args:
             axis (int or list of int): Axes along which the process groups are created.
             indices_at_axis (Optional[List[int]], optional): Indices at the axis. Defaults to None.
             backend (Optional[str], optional): Backend of the process group. Defaults to None.
+            return_ranks_by_group (bool): Whether to return all ranks by group for creating submesh. Defaults to False.
 
         Returns:
             ProcessGroup: The process group along the given axis which the current process belongs to.
@@ -267,6 +279,10 @@ def get_group_along_axis(
 
         coords_in_group = ProcessGroupMesh.get_coords_along_axis(self._coord, axis, indices_at_axis)
         ranks_in_group = tuple([ProcessGroupMesh.ravel(coord, self._shape) for coord in coords_in_group])
+
+        if return_ranks_by_group:
+            return self.create_group_along_axis(axis, indices_at_axis, backend=backend, return_ranks_by_group=True)
+
         if ranks_in_group not in self._ranks_to_group:
             # no need to cache it explicitly, since it will be cached in `create_group_along_axis`
             return self.create_group_along_axis(axis, indices_at_axis, backend=backend)

tests/test_shardformer/test_model/test_shard_mixtral.py

@@ -29,10 +29,10 @@ def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn,
     org_model, org_optimizer, sharded_model, sharded_optimizer, criterion, booster = build_model_from_hybrid_plugin(
         model_fn, loss_fn, test_config, pluggin_cls=MoeHybridParallelPlugin, optim_class=torch.optim.Adam
     )
-    with torch.autograd.set_detect_anomaly(True):
-        org_loss, org_output, sharded_loss, sharded_output = run_forward_backward_with_hybrid_plugin(
-            org_model, sharded_model, sharded_optimizer, data_gen_fn, output_transform_fn, criterion, booster
-        )
+
+    org_loss, org_output, sharded_loss, sharded_output = run_forward_backward_with_hybrid_plugin(
+        org_model, sharded_model, sharded_optimizer, data_gen_fn, output_transform_fn, criterion, booster
+    )
 
     stage_manager = booster.plugin.stage_manager
     tp_group = booster.plugin.tp_group
@@ -115,8 +115,9 @@ def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn,
     [
         {
             "tp_size": 1,
-            "pp_size": 1,
-            "ep_size": 1,
+            "pp_size": 2,
+            "num_microbatches": 2,
+            "ep_size": 2,
             "zero_stage": 1,
             "overlap_communication": False,
             "precision": "fp32",
@@ -125,7 +126,7 @@ def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn,
             "tp_size": 1,
             "pp_size": 2,
             "num_microbatches": 2,
-            "ep_size": 1,
+            "ep_size": 2,
             "zero_stage": 1,
             "overlap_communication": False,
             "precision": "fp32",
@@ -134,7 +135,7 @@ def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn,
             "tp_size": 2,
             "pp_size": 2,
             "num_microbatches": 2,
-            "ep_size": 1,
+            "ep_size": 2,
             "zero_stage": 1,
             "overlap_communication": False,
             "precision": "fp32",