Sync code and test CI; rename some env vars (#11686)

python/sglang/test/send_one.py

@@ -3,6 +3,8 @@ Run one test prompt.
 
 Usage:
 python3 -m sglang.test.send_one
+python3 -m sglang.test.send_one --profile --profile-steps 5
+python3 -m sglang.test.send_one --profile --profile-by-stage
 """
 
 import argparse
@@ -11,6 +13,8 @@ import json
 
 import requests
 
+from sglang.profiler import run_profile
+
 
 @dataclasses.dataclass
 class BenchArgs:
@@ -29,6 +33,9 @@ class BenchArgs:
     image: bool = False
     many_images: bool = False
     stream: bool = False
+    profile: bool = False
+    profile_steps: int = 3
+    profile_by_stage: bool = False
 
     @staticmethod
     def add_cli_args(parser: argparse.ArgumentParser):
@@ -51,6 +58,11 @@ class BenchArgs:
         parser.add_argument("--image", action="store_true")
         parser.add_argument("--many-images", action="store_true")
         parser.add_argument("--stream", action="store_true")
+        parser.add_argument("--profile", action="store_true")
+        parser.add_argument(
+            "--profile-steps", type=int, default=BenchArgs.profile_steps
+        )
+        parser.add_argument("--profile-by-stage", action="store_true")
 
     @classmethod
     def from_cli_args(cls, args: argparse.Namespace):
@@ -59,6 +71,8 @@ class BenchArgs:
 
 
 def send_one_prompt(args):
+    base_url = f"http://{args.host}:{args.port}"
+
     if args.image:
         args.prompt = (
             "Human: Describe this image in a very short sentence.\n\nAssistant:"
@@ -108,8 +122,20 @@ def send_one_prompt(args):
         "stream": args.stream,
     }
 
+    # Run profiler if requested
+    if args.profile:
+        print(f"Running profiler with {args.profile_steps} steps...")
+        run_profile(
+            base_url,
+            args.profile_steps,
+            ["CPU", "GPU"],
+            None,
+            None,
+            args.profile_by_stage,
+        )
+
     response = requests.post(
-        f"http://{args.host}:{args.port}/generate",
+        f"{base_url}/generate",
         json=json_data,
         stream=args.stream,
     )

python/sglang/test/test_utils.py

@@ -126,7 +126,7 @@ def is_in_ci():
 
 def is_in_amd_ci():
     """Return whether it is in an AMD CI runner."""
-    return get_bool_env_var("SGLANG_AMD_CI")
+    return get_bool_env_var("SGLANG_IS_IN_CI_AMD")
 
 
 def _use_cached_default_models(model_repo: str):