Added TPExpert for special situation

colossalai/nn/layer/moe/__init__.py

@@ -1,5 +1,8 @@
-from .experts import Experts, FFNExperts
+from .experts import Experts, FFNExperts, TPExperts
 from .layers import MoeLayer, Top1Router, Top2Router
-from .utils import NormalNoiseGenerator
+from .utils import NormalNoiseGenerator, build_ffn_experts
 
-__all__ = ['Experts', 'FFNExperts', 'Top1Router', 'Top2Router', 'MoeLayer', 'NormalNoiseGenerator']
+__all__ = [
+    'Experts', 'FFNExperts', 'TPExperts', 'Top1Router', 'Top2Router', 'MoeLayer', 'NormalNoiseGenerator',
+    'build_ffn_experts'
+]

colossalai/nn/layer/moe/_operation.py

@@ -15,6 +15,55 @@ except ImportError:
     print("If you want to activate cuda mode for MoE, please install with cuda_ext!")
 
 
+class AllGather(torch.autograd.Function):
+
+    @staticmethod
+    def forward(ctx: Any, inputs: Tensor, parallel_mode: ParallelMode) -> Tensor:
+
+        if ctx is not None:
+            ctx.parallel_mode = parallel_mode
+
+        comm_size = gpc.get_world_size(parallel_mode)
+        if comm_size == 1:
+            return inputs.unsqueeze(0)
+
+        buffer_shape = (comm_size,) + inputs.shape
+        outputs = torch.empty(buffer_shape, dtype=inputs.dtype, device=inputs.device)
+        buffer_list = list(torch.chunk(outputs, comm_size, dim=0))
+        dist.all_gather(buffer_list, inputs, group=gpc.get_group(parallel_mode))
+        return outputs
+
+    @staticmethod
+    def backward(ctx: Any, grad_outputs: Tensor) -> Tuple[Tensor, None]:
+        return ReduceScatter.forward(None, grad_outputs, ctx.parallel_mode), None
+
+
+class ReduceScatter(torch.autograd.Function):
+
+    @staticmethod
+    def forward(ctx: Any, inputs: Tensor, parallel_mode: ParallelMode) -> Tensor:
+
+        if ctx is not None:
+            ctx.parallel_mode = parallel_mode
+
+        comm_size = gpc.get_world_size(parallel_mode)
+        if comm_size == 1:
+            return inputs.squeeze(0)
+
+        if not inputs.is_contiguous():
+            inputs = inputs.contiguous()
+
+        output_shape = inputs.shape[1:]
+        outputs = torch.empty(output_shape, dtype=inputs.dtype, device=inputs.device)
+        buffer_list = list(torch.chunk(inputs, comm_size, dim=0))
+        dist.reduce_scatter(outputs, buffer_list, group=gpc.get_group(parallel_mode))
+        return outputs
+
+    @staticmethod
+    def backward(ctx: Any, grad_outputs: Tensor) -> Tuple[Tensor, None]:
+        return AllGather.forward(None, grad_outputs, ctx.parallel_mode), None
+
+
 class AllToAll(torch.autograd.Function):
     """Dispatches input tensor [e, c, h] to all experts by all_to_all_single
     operation in torch.distributed.

colossalai/nn/layer/moe/experts.py

@@ -7,7 +7,16 @@ from colossalai.context import ParallelMode, seed
 from colossalai.utils import get_current_device
 
 
-class Experts(nn.Module):
+class MoeExperts(nn.Module):
+
+    def __init__(self, comm: str):
+        super().__init__()
+        assert comm in {"all_to_all", "all_gather"}, \
+            "This kind of communication has not been implemented yet.\n Please use Experts build function."
+        self.comm = comm
+
+
+class Experts(MoeExperts):
     """A wrapper class to create experts. It will create E experts across the
     moe model parallel group, where E is the number of experts. Every expert
     is a instence of the class, 'expert' in initialization parameters.
@@ -20,19 +29,22 @@ class Experts(nn.Module):
     """
 
     def __init__(self, expert, num_experts, **expert_args):
-        super().__init__()
+        super().__init__("all_to_all")
 
         assert num_experts % moe_env.model_parallel_size == 0, \
             "The number of experts should be divied by moe model size"
 
         num_local_experts = num_experts // moe_env.model_parallel_size
+
         with seed(ParallelMode.MOE_MODEL):
             self.experts = nn.ModuleList([expert(**expert_args) for _ in range(num_local_experts)])
-        self.num_local_experts = num_local_experts
+
         for exp in self.experts:
             for param in exp.parameters():
                 param.__setattr__('moe_param', True)
 
+        self.num_local_experts = num_local_experts
+
     def forward(self, inputs):
         expert_input = torch.chunk(inputs, self.num_local_experts, dim=1)
         expert_output = []
@@ -44,10 +56,10 @@ class Experts(nn.Module):
         return output
 
 
-class FFNExperts(nn.Module):
+class FFNExperts(MoeExperts):
 
     def __init__(self, num_experts: int, d_model: int, d_ff: int, activation=None, drop_rate: float = 0):
-        super().__init__()
+        super().__init__("all_to_all")
 
         assert num_experts % moe_env.model_parallel_size == 0, \
             "The number of experts should be divied by moe model size"
@@ -75,7 +87,7 @@ class FFNExperts(nn.Module):
         for param in self.parameters():
             param.__setattr__('moe_param', True)
 
-    def forward(self, inputs):    # x [g, el, c, h]
+    def forward(self, inputs):    # inputs [g, el, c, h]
 
         el = inputs.size(1)
         h = inputs.size(-1)
@@ -96,3 +108,58 @@ class FFNExperts(nn.Module):
         outputs = outputs.reshape(inshape)
         outputs = outputs.transpose(0, 1).contiguous()
         return outputs
+
+
+class TPExperts(MoeExperts):
+
+    def __init__(self, num_experts: int, d_model: int, d_ff: int, activation=None, drop_rate: float = 0):
+        super().__init__("all_gather")
+
+        assert d_ff % moe_env.model_parallel_size == 0, \
+            "d_ff should be divied by moe model size"
+
+        p_ff = d_ff // moe_env.model_parallel_size
+
+        self.w1 = nn.Parameter(torch.empty(num_experts, d_model, p_ff, device=get_current_device()))
+        self.b1 = nn.Parameter(torch.empty(num_experts, 1, p_ff, device=get_current_device()))
+
+        self.w2 = nn.Parameter(torch.empty(num_experts, p_ff, d_model, device=get_current_device()))
+        self.b2 = nn.Parameter(torch.empty(num_experts, 1, d_model, device=get_current_device()))
+
+        s1 = math.sqrt(0.1 / d_model)
+        s2 = math.sqrt(0.1 / d_ff)
+
+        with seed(ParallelMode.MOE_MODEL):
+            nn.init.trunc_normal_(self.w1, std=s1)
+            nn.init.trunc_normal_(self.b1, std=s1)
+            nn.init.trunc_normal_(self.w2, std=s2)
+
+        nn.init.trunc_normal_(self.b2, std=s2)
+
+        self.act = nn.GELU() if activation is None else activation
+        self.drop = nn.Dropout(p=drop_rate)
+
+        self.w1.__setattr__('moe_param', True)
+        self.w2.__setattr__('moe_param', True)
+        self.b1.__setattr__('moe_param', True)
+
+    def forward(self, inputs):    # inputs [g, e, c, h]
+
+        e = inputs.size(1)
+        h = inputs.size(-1)
+
+        inputs = inputs.transpose(0, 1)
+        inshape = inputs.shape
+        inputs = inputs.reshape(e, -1, h)
+
+        out_ff = torch.baddbmm(self.b1, inputs, self.w1)
+        out_act = self.act(out_ff)
+        with seed(ParallelMode.TENSOR):
+            inter = self.drop(out_act)
+
+        out_model = torch.baddbmm(self.b2, inter, self.w2)
+        outputs = self.drop(out_model)    # outputs [e, gc, h]
+
+        outputs = outputs.reshape(inshape)
+        outputs = outputs.transpose(0, 1).contiguous()
+        return outputs    # outputs [g, e, c, h]

colossalai/nn/layer/moe/layers.py

@@ -8,7 +8,8 @@ from colossalai.core import global_context as gpc
 from colossalai.global_variables import moe_env
 from colossalai.context import ParallelMode
 from colossalai.utils import get_current_device
-from ._operation import U_CUDA_MODE, AllToAll, MoeDispatch, MoeCombine, moe_cumsum
+from ._operation import U_CUDA_MODE, AllToAll, AllGather, ReduceScatter, MoeDispatch, MoeCombine, moe_cumsum
+from .experts import MoeExperts
 from .utils import autocast_softmax
 
 
@@ -198,7 +199,7 @@ class MoeLayer(nn.Module):
     :type experts: nn.Module
     """
 
-    def __init__(self, dim_model: int, num_experts: int, router: nn.Module, experts: nn.Module):
+    def __init__(self, dim_model: int, num_experts: int, router: nn.Module, experts: MoeExperts):
         super().__init__()
         self.d_model = dim_model
         self.num_experts = num_experts
@@ -207,8 +208,8 @@ class MoeLayer(nn.Module):
         self.experts = experts
         self.cuda_mode = True if U_CUDA_MODE and moe_env.enable_cuda else False
 
-    def expert_part(self, expert_input: torch.Tensor):
+    def a2a_process(self, dispatch_data: torch.Tensor):
-        expert_input = AllToAll.apply(expert_input, ParallelMode.MOE_MODEL)
+        expert_input = AllToAll.apply(dispatch_data, ParallelMode.MOE_MODEL)
 
         input_shape = expert_input.shape
 
@@ -221,24 +222,42 @@ class MoeLayer(nn.Module):
         expert_output = AllToAll.apply(expert_output, ParallelMode.MOE_MODEL)
         return expert_output
 
+    def tp_process(self, dispatch_data: torch.Tensor):
+        expert_in = AllGather.apply(dispatch_data, ParallelMode.MOE_MODEL)
+        expert_out = self.experts(expert_in)
+        expert_out = ReduceScatter.apply(expert_out, ParallelMode.MOE_MODEL)
+        return expert_out
+
     def forward(self, inputs: torch.Tensor) -> torch.Tensor:
         tokens = inputs.reshape(-1, self.d_model)
         gate_output = self.gate(tokens)
         router_res = self.router(gate_output, self.cuda_mode)
 
         if self.cuda_mode:
-            logits, mask, dest_idx, ec = router_res
+            dispatch_data = MoeDispatch.apply(tokens, *router_res[1:])
-            expert_input = MoeDispatch.apply(tokens, mask, dest_idx, ec)
+            dispatch_data = dispatch_data.reshape(self.num_experts, -1, self.d_model)
-            expert_output = self.expert_part(expert_input)
-            ret = MoeCombine.apply(expert_output, logits, mask, dest_idx, ec)
         else:
-            combine_weights, sec_mask = router_res
+            sec_mask_f = router_res[1].type_as(inputs)
-            sec_mask_f = sec_mask.type_as(inputs)
+            dispatch_data = torch.matmul(sec_mask_f.permute(1, 2, 0), tokens)
-            expert_input = torch.matmul(sec_mask_f.permute(1, 2, 0), tokens)
+
-            expert_output = self.expert_part(expert_input)
+        # dispatch_data [e, c, h]
+        if self.experts.comm == "all_to_all":
+            expert_output = self.a2a_process(dispatch_data)
+        elif self.experts.comm == "all_gather":
+            expert_output = self.tp_process(dispatch_data)
+        else:
+            raise NotImplementedError("This kind of communication has not been implemented yet.\n Please use Experts "
+                                      "build function.")
+        # expert_output [e, c, h]
+
+        if self.cuda_mode:
+            expert_output = expert_output.reshape(-1, self.d_model)
+            ans = MoeCombine.apply(expert_output, *router_res)
+        else:
+            combine_weights = router_res[0]
             combine_weights = combine_weights.view(combine_weights.shape[0], -1)
             expert_output = expert_output.view(-1, expert_output.shape[-1])
-            ret = torch.matmul(combine_weights, expert_output)
+            ans = torch.matmul(combine_weights, expert_output)
 
-        ret = ret.reshape(inputs.shape)
+        ans = ans.reshape(inputs.shape)
-        return ret
+        return ans

colossalai/nn/layer/moe/utils.py

@@ -1,6 +1,8 @@
 import torch
 import torch.nn.functional as F
 from colossalai.utils import get_current_device
+from colossalai.global_variables import moe_env
+from .experts import FFNExperts, TPExperts
 
 
 class NormalNoiseGenerator:
@@ -14,10 +16,9 @@ class NormalNoiseGenerator:
     """
 
     def __init__(self, num_experts: int):
-        self.normal = torch.distributions.normal.Normal(
+        self.normal = torch.distributions.normal.Normal(loc=torch.tensor(0.0, device=get_current_device()),
-            loc=torch.tensor(0.0, device=get_current_device()),
+                                                        scale=torch.tensor(1.0 / num_experts**2,
-            scale=torch.tensor(1.0 / num_experts ** 2, device=get_current_device())
+                                                                           device=get_current_device())).rsample
-        ).rsample
 
     def __call__(self, inputs: torch.Tensor):
         noisy = self.normal(inputs.shape)
@@ -30,3 +31,13 @@ def autocast_softmax(inputs: torch.Tensor, dim: int):
     sm_input = inputs.to(torch.float32) if fp16_flag else inputs
     sm_output = F.softmax(sm_input, dim)
     return sm_output
+
+
+def build_ffn_experts(num_experts: int, d_model: int, d_ff: int, activation=None, drop_rate: float = 0):
+    moe_mp_size = moe_env.model_parallel_size
+    if num_experts % moe_mp_size == 0:
+        return FFNExperts(num_experts, d_model, d_ff, activation, drop_rate)
+    elif d_ff % moe_mp_size == 0:
+        return TPExperts(num_experts, d_model, d_ff, activation, drop_rate)
+    else:
+        raise NotImplementedError(f"Can not build {num_experts} experts in {moe_mp_size} GPUS.")

model_zoo/moe/models.py

@@ -4,7 +4,7 @@ import torch.nn as nn
 from colossalai.context import ParallelMode
 from colossalai.nn.layer import VanillaPatchEmbedding, VanillaClassifier, \
     WrappedDropout as Dropout, WrappedDropPath as DropPath
-from colossalai.nn.layer.moe import FFNExperts, MoeLayer, Top2Router, NormalNoiseGenerator
+from colossalai.nn.layer.moe import build_ffn_experts, MoeLayer, Top2Router, NormalNoiseGenerator
 from .util import moe_sa_args, moe_mlp_args
 from ..helper import TransformerLayer
 from colossalai.global_variables import moe_env
@@ -110,7 +110,7 @@ class Widenet(nn.Module):
 
         noisy_func = NormalNoiseGenerator(num_experts)
         shared_router = Top2Router(capacity_factor, noisy_func=noisy_func)
-        shared_experts = FFNExperts(num_experts, d_model, d_ff, drop_rate=drop_rate)
+        shared_experts = build_ffn_experts(num_experts, d_model, d_ff, drop_rate=drop_rate)
 
         # stochastic depth decay rule
         dpr = [x.item() for x in torch.linspace(0, drop_path, depth)]
@@ -177,7 +177,7 @@ class ViTMoE(nn.Module):
             ffn = VanillaFFN(**moe_mlp_args(
                 d_model=d_model, d_ff=d_ff, drop_rate=drop_rate)) if i % 2 == 0 else \
                 MoeLayer(dim_model=d_model, num_experts=num_experts, router=router,
-                         experts=FFNExperts(num_experts, d_model, d_ff, drop_rate=drop_rate))
+                         experts=build_ffn_experts(num_experts, d_model, d_ff, drop_rate=drop_rate))
             layer = TransformerLayer(att=sa,
                                      ffn=ffn,
                                      norm1=nn.LayerNorm(d_model, eps=1e-6),

tests/test_moe/short_test.py

@@ -1,6 +1,4 @@
-import os
 from functools import partial
-from pathlib import Path
 import pytest
 import torch
 import torch.nn as nn
@@ -9,10 +7,10 @@ import colossalai
 from colossalai.context import ParallelMode
 from colossalai.core import global_context as gpc
 from colossalai.utils import free_port, get_current_device
-from colossalai.nn.layer.moe import Top2Router, MoeLayer
+from colossalai.nn.layer.moe import Top2Router, MoeLayer, Experts
+from colossalai.context.random import moe_set_seed
 from colossalai.global_variables import moe_env
 
-
 BATCH_SIZE = 32
 NUM_EXPERTS = 4
 CONFIG = dict(parallel=dict(moe=dict(size=4)))
@@ -24,17 +22,17 @@ def check_equal(A, B, atol=1e-06):
 
 def run_routing(rank, world_size, port, rs=2, hidden_size=128, data_type=torch.float32):
     colossalai.launch(config=CONFIG, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
-
+    moe_set_seed(42)
     # torch.set_printoptions(precision=30)
     torch.backends.cuda.matmul.allow_tf32 = False
     local_rank = gpc.get_local_rank(ParallelMode.GLOBAL)
     torch.manual_seed(rs + local_rank)
     moe_env.reset_loss()
-    tokens = torch.randn(BATCH_SIZE, hidden_size,
+    tokens = torch.randn(BATCH_SIZE, hidden_size, dtype=data_type, device=get_current_device(), requires_grad=True)
-                         dtype=data_type, device=get_current_device(), requires_grad=True)
     # print(f"tokens:\n{tokens}")
     router = Top2Router(1)
-    layer = MoeLayer(hidden_size, NUM_EXPERTS, router, nn.Identity())
+    expert = Experts(nn.Identity, 4)
+    layer = MoeLayer(hidden_size, NUM_EXPERTS, router, expert)
     if data_type == torch.float16:
         layer = layer.half()
     layer.cuda_mode = False
@@ -88,8 +86,12 @@ def run_routing(rank, world_size, port, rs=2, hidden_size=128, data_type=torch.f
 @pytest.mark.parametrize("data_type", [torch.float32, torch.float16])
 def test_moe_top2(rs, hidden_size, data_type):
     world_size = 4
-    run_func = partial(run_routing, world_size=world_size, port=free_port(),
+    run_func = partial(run_routing,
-                       rs=rs, hidden_size=hidden_size, data_type=data_type)
+                       world_size=world_size,
+                       port=free_port(),
+                       rs=rs,
+                       hidden_size=hidden_size,
+                       data_type=data_type)
     mp.spawn(run_func, nprocs=world_size)