Use public model for FA3 speculative decode testing (sgl-project#5152)

Author: yubofredwang

test/srt/test_fa3.py

@@ -7,8 +7,6 @@
 from sglang.srt.utils import get_device_sm, kill_process_tree
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
-    DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
-    DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
     DEFAULT_MLA_MODEL_NAME_FOR_TEST,
     DEFAULT_MODEL_NAME_FOR_TEST,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -125,7 +123,7 @@ def get_server_args(cls):
 class TestFlashAttention3SpeculativeDecode(BaseFlashAttentionTest):
     """Test FlashAttention3 with speculative decode enabled."""
 
-    model = DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST
+    model = "meta-llama/Llama-3.1-8B-Instruct"
 
     @classmethod
     def get_server_args(cls):
@@ -137,7 +135,7 @@ def get_server_args(cls):
                 "--speculative-algorithm",
                 "EAGLE3",
                 "--speculative-draft",
-                DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
+                "jamesliu1/sglang-EAGLE3-Llama-3.1-Instruct-8B",
                 "--speculative-num-steps",
                 "3",
                 "--speculative-eagle-topk",