[FIX] Fix the nightly CI by disabling swa mem pool for gemma2 (#8693)

python/sglang/srt/server_args.py

@@ -1928,6 +1928,12 @@ class ServerArgs:
         if "Llama4" in model_arch:
             assert self.attention_backend == "fa3", "fa3 is required for Llama4 model"
 
+        if "Gemma2ForCausalLM" in model_arch:
+            # FIXME: https://github.com/sgl-project/sglang/pull/7367 is not compatible with gemma2 model.
+            # It failed at this test: https://github.com/sgl-project/sglang/actions/runs/16255155597/job/45890331952#step:4:736
+            logger.warning("Disable hybrid SWA memory for Gemma2ForCausalLM.")
+            self.disable_hybrid_swa_memory = True
+
         # Check LoRA
         self.check_lora_server_args()
 

python/sglang/test/test_utils.py

@@ -27,9 +27,6 @@ import torch.nn.functional as F
 
 from sglang.bench_serving import run_benchmark
 from sglang.global_config import global_config
-from sglang.lang.backend.openai import OpenAI
-from sglang.lang.backend.runtime_endpoint import RuntimeEndpoint
-from sglang.lang.interpreter import ProgramState
 from sglang.srt.utils import (
     get_bool_env_var,
     get_device,
@@ -358,6 +355,9 @@ def add_common_sglang_args_and_parse(parser: argparse.ArgumentParser):
 
 
 def select_sglang_backend(args: argparse.Namespace):
+    from sglang.lang.backend.openai import OpenAI
+    from sglang.lang.backend.runtime_endpoint import RuntimeEndpoint
+
     if args.backend.startswith("srt"):
         if args.backend == "srt-no-parallel":
             global_config.enable_parallel_encoding = False