[ci] fix ci test fused_moe op (#5102)

test/srt/run_suite.py

@@ -76,6 +76,7 @@ suites = {
         TestFile("test_create_kvindices.py", 2),
         TestFile("test_hicache.py", 60),
         TestFile("test_hicache_mla.py", 90),
+        TestFile("test_fused_moe.py", 30),
         TestFile("test_triton_moe_channel_fp8_kernel.py", 25),
     ],
     "per-commit-2-gpu": [

test/srt/test_fused_moe.py

@@ -3,7 +3,6 @@ import unittest
 import torch
 import torch.nn.functional as F
 from tqdm import tqdm
-from vllm.model_executor.layers.fused_moe import fused_moe as fused_moe_vllm
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
@@ -45,7 +44,18 @@ class TestFusedMOE(CustomTestCase):
         else:
             return 1e-2, 1e-2  # Default values for other types
 
-    def torch_naive_moe(self, a, w1, w2, score, topk):
+    def torch_naive_moe(
+        self,
+        a,
+        w1,
+        w2,
+        score,
+        topk,
+        w1_scale=None,
+        w2_scale=None,
+        a1_scale=None,
+        a2_scale=None,
+    ):
         B, D = a.shape
         a = a.view(B, -1, D).repeat(1, topk, 1).reshape(-1, D)
         out = torch.zeros(B * topk, w2.shape[1], dtype=a.dtype, device=a.device)
@@ -53,12 +63,30 @@ class TestFusedMOE(CustomTestCase):
         topk_weight, topk_ids = torch.topk(score, topk)
         topk_weight = topk_weight.view(-1)
         topk_ids = topk_ids.view(-1)
-        for i in range(w1.shape[0]):
+
+        if w1.dtype == torch.float8_e4m3fn:
+            w1_compute = w1.to(a.dtype)
+            w2_compute = w2.to(a.dtype)
+
+            if w1_scale is not None:
+                w1_compute = (w1_compute * w1_scale.view(-1, 1, 1)).to(a.dtype)
+            if w2_scale is not None:
+                w2_compute = (w2_compute * w2_scale.view(-1, 1, 1)).to(a.dtype)
+            if a1_scale is not None:
+                a = (a * a1_scale).to(a.dtype)
+            if a2_scale is not None:
+                a = (a * a2_scale).to(a.dtype)
+        else:
+            w1_compute = w1
+            w2_compute = w2
+
+        for i in range(w1_compute.shape[0]):
             mask = topk_ids == i
             if mask.sum():
-                out[mask] = SiluAndMul()(a[mask] @ w1[i].transpose(0, 1)) @ w2[
-                    i
-                ].transpose(0, 1)
+                out[mask] = SiluAndMul()(
+                    a[mask] @ w1_compute[i].transpose(0, 1)
+                ) @ w2_compute[i].transpose(0, 1)
+
         return (
             out.view(B, -1, w2.shape[1]) * topk_weight.view(B, -1, 1).to(out.dtype)
         ).sum(dim=1)
@@ -98,21 +126,12 @@ class TestFusedMOE(CustomTestCase):
                 a2_scale=a2_scale,
             )
 
-            vllm_output = fused_moe_vllm(
-                a,
-                w1,
-                w2,
-                score,
-                topk,
-                renormalize=False,
-                use_fp8_w8a8=True,
-                w1_scale=w1_scale,
-                w2_scale=w2_scale,
-                a1_scale=a1_scale,
-                a2_scale=a2_scale,
+            torch_output = self.torch_naive_moe(
+                a, w1, w2, score, topk, w1_scale, w2_scale, a1_scale, a2_scale
+            )
+            torch.testing.assert_close(
+                sglang_output, torch_output, rtol=rtol, atol=atol
             )
-
-            torch.testing.assert_close(sglang_output, vllm_output, rtol=rtol, atol=atol)
 
         else:
             a = self.create_random_cuda_tensor((m, k), dtype)
@@ -127,8 +146,8 @@ class TestFusedMOE(CustomTestCase):
             )
 
     def test_various_configurations(self):
-        m_values = [1, 33, 64, 222, 1024 * 128]
-        n_values = [128, 1024, 2048]
+        m_values = [1, 33, 64, 222]
+        n_values = [128, 1024]
         k_values = [128, 511, 1024]
         dtypes = [torch.float16, torch.bfloat16]
         fp8_modes = [False, True]
@@ -171,6 +190,7 @@ class TestFusedMOE(CustomTestCase):
                                                 dtype,
                                                 use_fp8_w8a8=use_fp8_w8a8,
                                             )
+                                            torch.cuda.empty_cache()
                                         pbar.update(1)