fix flashinfer & http log level

docs/release_process.md

@@ -1,3 +1,10 @@
+# PyPI Package Release Process
+
+## Update the version in code
+Update the package version in `python/pyproject.toml` and `python/sglang/__init__.py`.
+
+## Upload the PyPI package
+
 ```
 pip install build twine
 ```
@@ -5,4 +12,7 @@ pip install build twine
 ```
 cd python
 bash upload_pypi.sh
-```
+```
+
+## Make a release in GitHub
+Make a new release https://github.com/sgl-project/sglang/releases/new.

python/sglang/srt/server.py

@@ -152,7 +152,9 @@ def launch_server(server_args: ServerArgs, pipe_finish_writer, model_overide_arg
     if server_args.disable_disk_cache:
         disable_cache()
     if not server_args.disable_flashinfer:
-        assert_pkg_version("flashinfer", "0.0.8")
+        assert_pkg_version("flashinfer", "0.0.8", "Please uninstall the old version and "
+                           "reinstall the latest version by following the instructions "
+                           "at https://docs.flashinfer.ai/installation.html.")
     if server_args.chat_template:
         # TODO: replace this with huggingface transformers template
         load_chat_template_for_openai_api(server_args.chat_template)
@@ -293,7 +295,7 @@ def launch_server(server_args: ServerArgs, pipe_finish_writer, model_overide_arg
             app,
             host=server_args.host,
             port=server_args.port,
-            log_level=server_args.log_level_http or server_args.log_level_http,
+            log_level=server_args.log_level_http or server_args.log_level,
             timeout_keep_alive=5,
             loop="uvloop",
         )

python/sglang/srt/server_args.py

@@ -50,11 +50,11 @@ class ServerArgs:
     load_balance_method: str = "round_robin"
 
     # Optimization/debug options
-    disable_flashinfer: bool = True
+    disable_flashinfer: bool = False
-    attention_reduce_in_fp32: bool = False
     disable_radix_cache: bool = False
     disable_regex_jump_forward: bool = False
     disable_disk_cache: bool = False
+    attention_reduce_in_fp32: bool = False
 
     # Distributed args
     nccl_init_addr: Optional[str] = None

python/sglang/srt/utils.py

@@ -426,17 +426,19 @@ def suppress_other_loggers():
     logging.getLogger("vllm.utils").setLevel(logging.WARN)
 
 
-def assert_pkg_version(pkg: str, min_version: str):
+def assert_pkg_version(pkg: str, min_version: str, message: str):
     try:
         installed_version = version(pkg)
         if pkg_version.parse(installed_version) < pkg_version.parse(min_version):
             raise Exception(
-                f"{pkg} is installed with version {installed_version} which "
+                f"{pkg} is installed with version {installed_version}, which "
-                f"is less than the minimum required version {min_version}"
+                f"is less than the minimum required version {min_version}. " +
+                message
             )
     except PackageNotFoundError:
         raise Exception(
-            f"{pkg} with minimum required version {min_version} is not installed"
+            f"{pkg} with minimum required version {min_version} is not installed. " +
+            message
         )
 
 
@@ -588,4 +590,4 @@ def receive_addrs(model_port_args, server_args):
         print(f"Node 0 received from rank {src_rank}: {tensor.tolist()}")
 
     dist.barrier()
-    dist.destroy_process_group() 
+    dist.destroy_process_group()