[fix]fix tests of v1 and worker

tests/worker/test_model_runner.py

@@ -32,7 +32,7 @@ def test_deepseek_mla_attn_backend_module():
         trust_remote_code=True,
         enable_chunked_prefill=False,
     )
-    assert model_runner.attn_backend.__name__ == "TritonMLABackend"
+    assert model_runner.attn_backend.__name__ == "FlashMLABackend" # "TritonMLABackend"
 
 
 @pytest.mark.parametrize("batch_size", list(range(1, 257, 3)))

vllm/model_executor/layers/fused_moe/layer.py

@@ -799,7 +799,6 @@ class FusedMoE(torch.nn.Module):
         if quant_config is None:
             # Not considering quant for now, temporarily
             self.use_nn_moe = int(os.environ.get('MOE_NN', 1)) == 1
-            # self.use_nn_moe = os.environ.get('MOE_NN') == '1'
         else:
             self.use_nn_moe = False