[Refactor] Remove legacy stage_args YAML loader (#6200)

Signed-off-by: zwhzzz0821 <2831474076@qq.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>

examples/offline_inference/aura_omni/end2end.py

@@ -24,7 +24,6 @@ from vllm.assets.video import VideoAsset, video_to_ndarrays
 from vllm.multimodal.media.audio import load_audio
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 
-from vllm_omni.engine.arg_utils import nullify_stage_engine_defaults
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.model_executor.stage_input_processors.aura_omni import (
     DEFAULT_QWEN3_TTS_REF_TEXT,
@@ -228,7 +227,6 @@ def parse_args():
     parser.add_argument("--init-timeout", type=int, default=2000, help="Overall initialization timeout.")
     parser.add_argument("--stage-init-timeout", type=int, default=2000, help="Per-stage initialization timeout.")
     parser.add_argument("--log-stats", action="store_true", default=False, help="Enable stats logging.")
-    nullify_stage_engine_defaults(parser)
     return parser.parse_args()
 
 

examples/offline_inference/lance/end2end.py

@@ -82,9 +82,6 @@ def parse_args():
     # ``max_num_seqs=1`` …) are passed as flat kwargs to ``Omni`` below
     # and ``create_default_diffusion`` materializes the stage config.
 
-    from vllm_omni.engine.arg_utils import nullify_stage_engine_defaults
-
-    nullify_stage_engine_defaults(p)
     return p.parse_args()
 
 

examples/offline_inference/nemotron_voicechat/end2end.py

@@ -67,7 +67,7 @@ def main() -> None:
     parser.add_argument("--output-dir", default="results/nemotron_voicechat")
     parser.add_argument("--system-prompt", default=DEFAULT_SYSTEM_PROMPT)
     parser.add_argument(
-        "--stage-configs-path",
+        "--deploy-config",
         default=None,
         help=(
             "Deploy yaml override. Pass vllm_omni/deploy/nemotron_labs_voicechat_streaming.yaml "
@@ -134,8 +134,8 @@ def main() -> None:
     ]
 
     omni_kwargs: dict = {}
-    if args.stage_configs_path:
-        omni_kwargs["stage_configs_path"] = args.stage_configs_path
+    if args.deploy_config:
+        omni_kwargs["deploy_config"] = args.deploy_config
     omni = Omni(model=str(ckpt), trust_remote_code=True, **omni_kwargs)
     outputs = omni.generate(inputs, sampling_params_list)
 

examples/offline_inference/text_to_speech/fish_speech/end2end.py

@@ -38,17 +38,6 @@ from vllm_omni.utils.tracking_parser import TrackingArgumentParser
 logger = logging.getLogger(__name__)
 
 DEFAULT_MODEL = "fishaudio/s2-pro"
-DEFAULT_DEPLOY_CONFIG = os.path.join(
-    os.path.dirname(__file__),
-    "..",
-    "..",
-    "..",
-    "..",
-    "vllm_omni",
-    "model_executor",
-    "stage_configs",
-    "fish_speech_s2_pro.yaml",
-)
 
 
 def build_prompt(
@@ -152,7 +141,8 @@ def _build_omni_kwargs(args, model_name: str) -> dict:
         log_stats=args.log_stats,
         stage_init_timeout=args.stage_init_timeout,
     )
-    omni_kwargs["deploy_config"] = args.deploy_config or DEFAULT_DEPLOY_CONFIG
+    if args.deploy_config:
+        omni_kwargs["deploy_config"] = args.deploy_config
     return omni_kwargs
 
 

examples/offline_inference/text_to_speech/qwen3_tts/word_timestamps.py

@@ -32,11 +32,10 @@ from end2end import _estimate_prompt_len
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 
 from vllm_omni import Omni
-from vllm_omni.engine.arg_utils import nullify_stage_engine_defaults
 from vllm_omni.utils.forced_aligner import align, build_forced_aligner_config
 
 
-def _default_stage_config() -> str:
+def _default_deploy_config() -> str:
     repo_root = Path(__file__).resolve().parents[4]
     return str(repo_root / "vllm_omni" / "deploy" / "qwen3_tts.yaml")
 
@@ -109,7 +108,7 @@ def main(args: Any) -> None:
         "output_dir",
     ):
         omni_kwargs.pop(key, None)
-    omni_kwargs["stage_configs_path"] = args.stage_configs_path or _default_stage_config()
+    omni_kwargs["deploy_config"] = args.deploy_config or _default_deploy_config()
     omni_kwargs["log_stats"] = args.log_stats
     omni = Omni(**omni_kwargs)
 
@@ -156,7 +155,7 @@ def parse_args() -> Any:
         default=None,
         help="Optional YAML file for forced aligner settings (incl. gpu_memory_utilization)",
     )
-    parser.add_argument("--stage-configs-path", default=None, help="Qwen3-TTS deploy YAML")
+    parser.add_argument("--deploy-config", default=None, help="Qwen3-TTS deploy YAML")
     parser.add_argument("--text", default="Hello world.", help="Text to synthesize and align")
     parser.add_argument("--language", default="English", help="Qwen3-TTS language field")
     parser.add_argument("--speaker", default="Vivian", help="CustomVoice speaker name")
@@ -164,7 +163,6 @@ def parse_args() -> Any:
     parser.add_argument("--max-new-tokens", type=int, default=2048, help="TTS max_new_tokens")
     parser.add_argument("--output-dir", default="output_audio", help="Directory for WAV and JSON sidecar")
     parser.add_argument("--log-stats", action="store_true", default=False, help="Enable vLLM-Omni stats logging")
-    nullify_stage_engine_defaults(parser)
     return parser.parse_args()