From 25e678d93336f089b7248b414c79ef66ad51302f Mon Sep 17 00:00:00 2001
From: Kangyan-Zhou <zky314343421@gmail.com>
Date: Fri, 6 Mar 2026 05:36:13 -0800
Subject: [PATCH] [diffusion] endpoint: add /server_info and /model_info
 endpoints for gateway discovery (#20020)

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---
 .../runtime/entrypoints/http_server.py        | 58 +++++++++++++++++++
 1 file changed, 58 insertions(+)

diff --git a/python/sglang/multimodal_gen/runtime/entrypoints/http_server.py b/python/sglang/multimodal_gen/runtime/entrypoints/http_server.py
index 29aeaa0f0..4830411bf 100644
--- a/python/sglang/multimodal_gen/runtime/entrypoints/http_server.py
+++ b/python/sglang/multimodal_gen/runtime/entrypoints/http_server.py
@@ -25,6 +25,7 @@ from sglang.multimodal_gen.runtime.entrypoints.utils import (
 from sglang.multimodal_gen.runtime.scheduler_client import async_scheduler_client
 from sglang.multimodal_gen.runtime.server_args import ServerArgs, get_global_server_args
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
+from sglang.version import __version__
 
 if TYPE_CHECKING:
     from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
@@ -95,6 +96,63 @@ async def get_models(request: Request):
     return response
 
 
+@health_router.get("/server_info")
+async def server_info_endpoint(request: Request):
+    """Get server information.
+
+    Returns fields compatible with the LLM engine's /server_info so that
+    the model gateway can discover diffusion workers.
+    """
+    server_args: ServerArgs = request.app.state.server_args
+
+    return {
+        "model_path": server_args.model_path,
+        "served_model_name": server_args.model_id or server_args.model_path,
+        "tp_size": server_args.tp_size,
+        "dp_size": server_args.dp_size,
+        "version": __version__,
+    }
+
+
+@health_router.get("/model_info")
+async def model_info_endpoint(request: Request):
+    """Get model information.
+
+    Returns fields compatible with the LLM engine's /model_info so that
+    the model gateway can detect capabilities for diffusion workers.
+    """
+    from sglang.multimodal_gen.registry import get_model_info
+
+    server_args: ServerArgs = request.app.state.server_args
+    task_type = server_args.pipeline_config.task_type
+
+    try:
+        registry_info = get_model_info(
+            server_args.model_path,
+            backend=server_args.backend,
+            model_id=server_args.model_id,
+        )
+    except Exception:
+        logger.warning("Failed to resolve model info from registry", exc_info=True)
+        registry_info = None
+
+    return {
+        # Fields consumed by the model gateway for worker discovery
+        "model_path": server_args.model_path,
+        "is_generation": True,
+        "model_type": "diffusion",
+        "architectures": (
+            [registry_info.pipeline_cls.__name__] if registry_info else None
+        ),
+        # Fields matching the LLM engine's /model_info shape
+        "has_image_understanding": task_type.accepts_image_input(),
+        "has_audio_understanding": False,
+        # Diffusion-specific fields
+        "task_type": task_type.name,
+        "is_image_gen": task_type.is_image_gen(),
+    }
+
+
 @health_router.get("/health_generate")
 async def health_generate():
     # TODO : health generate endpoint