MoE torch compile (#1497)

python/sglang/srt/layers/fused_moe/patch.py

@@ -0,0 +1,117 @@
+from typing import Optional
+
+import torch
+from torch.nn import functional as F
+
+
+def fused_topk_native(
+    hidden_states: torch.Tensor,
+    gating_output: torch.Tensor,
+    topk: int,
+    renormalize: bool,
+):
+    assert hidden_states.shape[0] == gating_output.shape[0], "Number of tokens mismatch"
+    M, _ = hidden_states.shape
+    topk_weights = torch.empty(
+        M, topk, dtype=torch.float32, device=hidden_states.device
+    )
+    topk_ids = torch.empty(M, topk, dtype=torch.int32, device=hidden_states.device)
+    topk_weights = F.softmax(gating_output.float(), dim=-1)
+    topk_weights, topk_ids = torch.topk(topk_weights, topk, dim=-1)
+    if renormalize:
+        topk_weights = topk_weights / topk_weights.sum(dim=-1, keepdim=True)
+    return topk_weights, topk_ids
+
+
+# This is used by the Deepseek-V2 model
+def grouped_topk(
+    hidden_states: torch.Tensor,
+    gating_output: torch.Tensor,
+    topk: int,
+    renormalize: bool,
+    num_expert_group: int = 0,
+    topk_group: int = 0,
+):
+
+    assert hidden_states.shape[0] == gating_output.shape[0], "Number of tokens mismatch"
+
+    scores = torch.softmax(gating_output, dim=-1)
+    num_token = scores.shape[0]
+    group_scores = (
+        scores.view(num_token, num_expert_group, -1).max(dim=-1).values
+    )  # [n, n_group]
+    group_idx = torch.topk(group_scores, k=topk_group, dim=-1, sorted=False)[
+        1
+    ]  # [n, top_k_group]
+    group_mask = torch.zeros_like(group_scores)  # [n, n_group]
+    group_mask.scatter_(1, group_idx, 1)  # [n, n_group]
+    score_mask = (
+        group_mask.unsqueeze(-1)
+        .expand(num_token, num_expert_group, scores.shape[-1] // num_expert_group)
+        .reshape(num_token, -1)
+    )  # [n, e]
+    tmp_scores = scores.masked_fill(~score_mask.bool(), 0.0)  # [n, e]
+    topk_weights, topk_ids = torch.topk(tmp_scores, k=topk, dim=-1, sorted=False)
+
+    if renormalize:
+        topk_weights = topk_weights / topk_weights.sum(dim=-1, keepdim=True)
+    return topk_weights, topk_ids
+
+
+def select_experts_native(
+    hidden_states: torch.Tensor,
+    router_logits: torch.Tensor,
+    top_k: int,
+    use_grouped_topk: bool,
+    renormalize: bool,
+    topk_group: Optional[int] = None,
+    num_expert_group: Optional[int] = None,
+):
+    # DeekSeekv2 uses grouped_top_k
+    if use_grouped_topk:
+        assert topk_group is not None
+        assert num_expert_group is not None
+        topk_weights, topk_ids = grouped_topk(
+            hidden_states=hidden_states,
+            gating_output=router_logits,
+            topk=top_k,
+            renormalize=renormalize,
+            num_expert_group=num_expert_group,
+            topk_group=topk_group,
+        )
+    else:
+        topk_weights, topk_ids = fused_topk_native(
+            hidden_states=hidden_states,
+            gating_output=router_logits,
+            topk=top_k,
+            renormalize=renormalize,
+        )
+    return topk_weights, topk_ids
+
+
+def fused_moe_forward_native(
+    layer: torch.nn.Module,
+    x: torch.Tensor,
+    use_grouped_topk: bool,
+    top_k: int,
+    router_logits: torch.Tensor,
+    renormalize: bool,
+    topk_group: Optional[int] = None,
+    num_expert_group: Optional[int] = None,
+) -> torch.Tensor:
+    topk_weights, topk_ids = select_experts_native(
+        hidden_states=x,
+        router_logits=router_logits,
+        use_grouped_topk=use_grouped_topk,
+        top_k=top_k,
+        renormalize=renormalize,
+        topk_group=topk_group,
+        num_expert_group=num_expert_group,
+    )
+    w13_weights = layer.w13_weight[topk_ids]
+    w1_weights, w3_weights = torch.chunk(w13_weights, 2, dim=2)
+    w2_weights = layer.w2_weight[topk_ids]
+    x1 = F.silu(torch.einsum("ti,taoi -> tao", x, w1_weights))
+    x3 = torch.einsum("ti, taoi -> tao", x, w3_weights)
+    expert_outs = torch.einsum("tao, taio -> tai", (x1 * x3), w2_weights)
+    return torch.einsum("tai,ta -> ti", expert_outs, topk_weights)

python/sglang/srt/model_executor/cuda_graph_runner.py

@@ -25,6 +25,7 @@ import torch
 from vllm.distributed.parallel_state import graph_capture
 from vllm.model_executor.custom_op import CustomOp
 
+from sglang.srt.layers.fused_moe.patch import fused_moe_forward_native
 from sglang.srt.layers.logits_processor import (
     LogitsMetadata,
     LogitsProcessor,
@@ -41,14 +42,15 @@ if TYPE_CHECKING:
 def _to_torch(model: torch.nn.Module, reverse: bool = False):
     for sub in model._modules.values():
         if isinstance(sub, CustomOp):
-            # NOTE: FusedMoE torch native implementaiton is not efficient
-            if "FusedMoE" in sub.__class__.__name__:
-                continue
             if reverse:
                 sub._forward_method = sub.forward_cuda
                 setattr(sub, "is_torch_compile", False)
             else:
-                sub._forward_method = sub.forward_native
+                # NOTE: Temporarily workaround MoE
+                if "FusedMoE" in sub.__class__.__name__:
+                    sub._forward_method = fused_moe_forward_native
+                else:
+                    sub._forward_method = sub.forward_native
                 setattr(sub, "is_torch_compile", True)
         if isinstance(sub, torch.nn.Module):
             _to_torch(sub, reverse)
@@ -67,7 +69,9 @@ def patch_model(
             monkey_patch_vllm_all_gather()
             backup_ca_comm = tp_group.ca_comm
             tp_group.ca_comm = None
-            yield torch.compile(model.forward, mode="max-autotune-no-cudagraphs")
+            yield torch.compile(
+                torch.no_grad()(model.forward), mode="max-autotune-no-cudagraphs"
+            )
         else:
             yield model.forward
     finally: