[Core] Composable parallel: Phase 1 strategy overlay (TP/DP/stage-replica) (RFC #4084) (#4281)

Signed-off-by: Kushan A. <kahmadian@nvidia.com>
Signed-off-by: kushanam <kushanam@users.noreply.github.com>
Signed-off-by: Kushan A <kahmadian@nvidia.com>
Signed-off-by: kahmadian <kahmadian@users.noreply.github.com>
Co-authored-by: kushanam <kushanam@users.noreply.github.com>
Co-authored-by: kahmadian <kahmadian@users.noreply.github.com>

vllm_omni/entrypoints/cli/serve.py

@@ -254,6 +254,15 @@ class OmniServeCommand(CLISubcommand):
             help="Path to a deploy config YAML (new format with stages/engine_args). "
             "Mutually exclusive with --stage-configs-path.",
         )
+        omni_config_group.add_argument(
+            "--strategy-config",
+            type=str,
+            default=None,
+            help="Path to a composable-parallel strategy.yaml. Only applies to "
+            "registry-based models (e.g. qwen2_5_omni): its derived parallel sizing "
+            "is overlaid onto the registry-merged stages before per-stage engine "
+            "args are built.",
+        )
         omni_config_group.add_argument(
             "--stage-overrides",
             type=str,
@@ -766,7 +775,10 @@ def run_headless(args: TrackingNamespace) -> None:
         load_omni_transfer_config_for_model,
         prepare_engine_environment,
     )
-    from vllm_omni.entrypoints.utils import load_and_resolve_stage_configs
+    from vllm_omni.entrypoints.utils import (
+        load_and_resolve_stage_configs,
+        parse_stage_overrides,
+    )
 
     model = args.model
     stage_id: int | None = args.stage_id
@@ -802,11 +814,18 @@ def run_headless(args: TrackingNamespace) -> None:
             args.replica_id,
         )
 
-    config_path, stage_configs = load_and_resolve_stage_configs(
+    # Parse --stage-overrides (raw JSON string) exactly like the standard
+    # engine path (AsyncOmniEngine._resolve_stage_configs) so headless and
+    # standard launches resolve to the same per-stage device layout.
+    stage_overrides = parse_stage_overrides(args_dict.get("stage_overrides"))
+
+    config_path, stage_configs, _ = load_and_resolve_stage_configs(
         model,
         stage_configs_path,
         args_dict,
         deploy_config_path=args_dict.get("deploy_config"),
+        stage_overrides=stage_overrides,
+        strategy_config_path=args_dict.get("strategy_config"),
     )
 
     # Locate the stage config that matches stage_id.