fix test

Signed-off-by: jiahanc <[email protected]>

Author: jiahanc

tests/moe/test_trtllm_gen_fused_moe.py

@@ -1923,7 +1923,7 @@ def cache_permute_indices():
         ),
         pytest.param(
             {
-                "num_experts": 512,
+                "num_experts": 128,
                 "top_k": 10,
                 "padding": 8,
                 "n_groups": None,
@@ -1937,7 +1937,7 @@ def cache_permute_indices():
         ),
         pytest.param(
             {
-                "num_experts": 256,
+                "num_experts": 128,
                 "top_k": 8,
                 "padding": 8,
                 "n_groups": None,
@@ -2062,14 +2062,6 @@ def test_moe_quantization_classes(
             f"Skipping for testing speed: intermediate_size={intermediate_size} with {routing_config['num_experts']} experts"
         )
 
-    # Skip large intermediate size and hidden size for configurations with small epxerts
-    if routing_config["num_experts"] < 512 and (
-        intermediate_size > 512 or hidden_size > 1024
-    ):
-        pytest.skip(
-            f"Skipping for testing speed: intermediate_size={intermediate_size} with {routing_config['num_experts']} experts"
-        )
-
     if type(moe_impl) not in routing_config["compatible_moe_impls"]:
         pytest.skip(
             f"Incompatible: {moe_impl.name} + {routing_config['routing_method_type'].name}"