[Feature] Support Stage Based Deployment CLI (#939)

Signed-off-by: wuhang <wuhang6@huawei.com>
Signed-off-by: princepride <wangzhipeng628@gmail.com>
Signed-off-by: wuhang <whlbx@hotmail.com>
Co-authored-by: 汪志鹏 <wangzhipeng628@gmail.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: 3 people

examples/online_serving/bagel/run_server_stage_cli.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+# Bagel multi-stage online serving startup script
+# Starts stage 0 as master with API server, and stage 1 in headless mode
+
+MODEL="${MODEL:-ByteDance-Seed/BAGEL-7B-MoT}"
+PORT="${PORT:-8091}"
+MASTER_ADDRESS="${MASTER_ADDRESS:-127.0.0.1}"
+MASTER_PORT="${MASTER_PORT:-8092}"
+STAGE_CONFIGS_PATH="$(dirname "$0")/../../../vllm_omni/model_executor/stage_configs/bagel.yaml"
+
+echo "Starting Bagel multi-stage server..."
+echo "Model: $MODEL"
+echo "API Port: $PORT"
+echo "Master Address: $MASTER_ADDRESS"
+echo "Master Port: $MASTER_PORT"
+echo "Stage Configs: $STAGE_CONFIGS_PATH"
+
+# Start stage 1 (DiT) in headless mode first
+echo "Starting Stage 1 (DiT) in headless mode..."
+vllm serve "$MODEL" --omni \
+    --stage-configs-path "$STAGE_CONFIGS_PATH" \
+    --stage-id 1 \
+    --headless \
+    -oma "$MASTER_ADDRESS" \
+    -omp "$MASTER_PORT" &
+
+# Start stage 0 (Thinker) as master with API server
+echo "Starting Stage 0 (Thinker) as master..."
+vllm serve "$MODEL" --omni \
+    --port "$PORT" \
+    --stage-configs-path "$STAGE_CONFIGS_PATH" \
+    --stage-id 0 \
+    -oma "$MASTER_ADDRESS" \
+    -omp "$MASTER_PORT"

requirements/common.txt

@@ -14,3 +14,4 @@ imageio[ffmpeg]>=2.37.2
 sox>=1.5.0
 prettytable>=3.8.0
 aenum==3.1.16
+pyzmq>=25.0.0

tests/entrypoints/test_async_omni_diffusion_config.py

@@ -3,21 +3,23 @@
 
 import pytest
 
-from vllm_omni.entrypoints import omni as omni_module
+from vllm_omni.entrypoints import utils as utils_module
 from vllm_omni.entrypoints.async_omni import AsyncOmni
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
+MODEL = "riverclouds/qwen_image_random"
+
 
 def test_default_stage_config_includes_cache_backend(monkeypatch):
     """Ensure cache_backend/cache_config are preserved in default diffusion stage."""
-    monkeypatch.setattr(omni_module, "load_stage_configs_from_model", lambda model, base_engine_args=None: [])
-    monkeypatch.setattr(omni_module, "resolve_model_config_path", lambda model: None)
+    monkeypatch.setattr(utils_module, "load_stage_configs_from_model", lambda model, base_engine_args=None: [])
+    monkeypatch.setattr(utils_module, "resolve_model_config_path", lambda model: None)
     monkeypatch.setattr(AsyncOmni, "_start_stages", lambda self, model: None)
     monkeypatch.setattr(AsyncOmni, "_wait_for_stages_ready", lambda self, timeout=0: None)
 
     omni = AsyncOmni(
-        model="dummy-model",
+        model=MODEL,
         cache_backend="cache_dit",
         cache_config='{"Fn_compute_blocks": 2}',
         vae_use_slicing=True,
@@ -41,13 +43,13 @@ def test_default_stage_config_includes_cache_backend(monkeypatch):
 
 def test_default_cache_config_used_when_missing(monkeypatch):
     """Ensure default cache_config is applied when cache_backend is set."""
-    monkeypatch.setattr(omni_module, "load_stage_configs_from_model", lambda model, base_engine_args=None: [])
-    monkeypatch.setattr(omni_module, "resolve_model_config_path", lambda model: None)
+    monkeypatch.setattr(utils_module, "load_stage_configs_from_model", lambda model, base_engine_args=None: [])
+    monkeypatch.setattr(utils_module, "resolve_model_config_path", lambda model: None)
     monkeypatch.setattr(AsyncOmni, "_start_stages", lambda self, model: None)
     monkeypatch.setattr(AsyncOmni, "_wait_for_stages_ready", lambda self, timeout=0: None)
 
     omni = AsyncOmni(
-        model="dummy-model",
+        model=MODEL,
         cache_backend="cache_dit",
     )
 
@@ -59,13 +61,13 @@ def test_default_cache_config_used_when_missing(monkeypatch):
 
 def test_default_stage_devices_from_sequence_parallel(monkeypatch):
     """Ensure devices list reflects sequence parallel size when no parallel_config is provided."""
-    monkeypatch.setattr(omni_module, "load_stage_configs_from_model", lambda model, base_engine_args=None: [])
-    monkeypatch.setattr(omni_module, "resolve_model_config_path", lambda model: None)
+    monkeypatch.setattr(utils_module, "load_stage_configs_from_model", lambda model, base_engine_args=None: [])
+    monkeypatch.setattr(utils_module, "resolve_model_config_path", lambda model: None)
     monkeypatch.setattr(AsyncOmni, "_start_stages", lambda self, model: None)
     monkeypatch.setattr(AsyncOmni, "_wait_for_stages_ready", lambda self, timeout=0: None)
 
     omni = AsyncOmni(
-        model="dummy-model",
+        model=MODEL,
         ulysses_degree=2,
         ring_degree=2,
     )