[Eagle] Remove the greedy branch and some redundant code (#4363)

Co-authored-by: Sehoon Kim <sehoon@x.ai>

test/srt/test_eagle_infer.py

@@ -122,8 +122,8 @@ class TestEAGLEEngine(unittest.TestCase):
 
     def _test_acc_length(self, engine):
         prompt = [
-            "Human: Give me a fully functional FastAPI server. Show the python code.\n\nAssistant:"
-        ] * 5
+            "Human: Give me a fully functional FastAPI server. Show the python code.\n\nAssistant:",
+        ] * 5  # test batched generation
         sampling_params = {"temperature": 0, "max_new_tokens": 512}
         output = engine.generate(prompt, sampling_params)
         output = output[0]

test/srt/test_mla_flashinfer.py

@@ -67,7 +67,7 @@ class TestFlashinferMLANoRagged(unittest.TestCase):
                     "--enable-torch-compile",
                     "--disable-cuda-graph",
                     "--cuda-graph-max-bs",
-                    "2",
+                    "4",
                     "--enable-flashinfer-mla",
                     "--flashinfer-mla-disable-ragged",
                 ]
@@ -109,7 +109,7 @@ class TestFlashinferMLAMTP(unittest.TestCase):
             other_args.extend(
                 [
                     "--cuda-graph-max-bs",
-                    "2",
+                    "4",
                     "--disable-radix",
                     "--enable-torch-compile",
                     "--torch-compile-max-bs",