[refactor]: remove the flat-flag entry point of the OpenAI API server

python -m fastvideo.entrypoints.openai.api_server parsed the hand-written
FastVideoArgs argparse flags, an inference input path next to the typed config
that fastvideo serve reads. It is removed; start the server with
fastvideo serve --config SERVE_CONFIG and dotted overrides such as
--server.port 8000. The GPU integration test starts the server that way, with
the same model, GPU count, and DiT offload.

FastVideoArgs.add_cli_args stays for the legacy training and preprocessing
scripts.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012M91wnVFmPEvJ9h39r5BH7

fastvideo/entrypoints/openai/api_server.py

@@ -174,31 +174,6 @@ def create_app(
     return app
 
 
-def _parse_args() -> tuple[FastVideoArgs, str, int, str]:
-    """Parse CLI arguments and return (FastVideoArgs, host, port, output_dir)"""
-    from fastvideo.utils import FlexibleArgumentParser
-
-    logger.warning("python -m fastvideo.entrypoints.openai.api_server and its flat flags are deprecated and will be "
-                   "removed; use `fastvideo serve --config SERVE_CONFIG` with dotted overrides such as "
-                   "`--generator.engine.num_gpus 2` instead.")
-    parser = FlexibleArgumentParser(description="FastVideo OpenAI-compatible API server")
-    parser.add_argument("--host", type=str, default=DEFAULT_HOST)
-    parser.add_argument("--port", type=int, default=DEFAULT_PORT)
-    parser.add_argument("--output-dir", type=str, default=DEFAULT_OUTPUT_DIR)
-    parser = FastVideoArgs.add_cli_args(parser)
-
-    args = parser.parse_args()
-    host = args.host
-    port = args.port
-    output_dir = args.output_dir
-
-    # Build FastVideoArgs from the remaining CLI args
-    excluded = {"host", "port", "output_dir", "subparser", "config", "dispatch_function"}
-    cli_kwargs = {k: v for k, v in vars(args).items() if k not in excluded and v is not None}
-    fastvideo_args = FastVideoArgs.from_kwargs(**cli_kwargs)
-    return fastvideo_args, host, port, output_dir
-
-
 def run_server(
     fastvideo_args: FastVideoArgs,
     host: str = DEFAULT_HOST,
@@ -229,8 +204,3 @@ def run_server(
         log_level="info",
         timeout_keep_alive=300,
     )
-
-
-if __name__ == "__main__":
-    fastvideo_args, host, port, output_dir = _parse_args()
-    run_server(fastvideo_args, host, port, output_dir=output_dir)

fastvideo/entrypoints/openai/state.py

@@ -1,8 +1,7 @@
 """Global server state shared across API modules.
 
-Keeping state in a dedicated module prevents the classic '__main__ vs package
-module' duplication that occurs when api_server.py is run with ``python -m``.
-All modules that need the generator or server args should import from here.
+Keeping state in a dedicated module gives every API module the same
+generator and server args. All modules that need them should import from here.
 """
 
 from __future__ import annotations

fastvideo/tests/entrypoints/test_openai_api_integration.py

@@ -59,21 +59,23 @@ def server():
     env.setdefault("MASTER_ADDR", "127.0.0.1")
     env.setdefault("MASTER_PORT", str(_find_free_port()))
 
+    config_path = os.path.join(output_dir, "serve.yaml")
+    with open(config_path, "w", encoding="utf-8") as config_file:
+        config_file.write(f"generator:\n  model_path: {MODEL_PATH}\n"
+                          "  engine:\n    num_gpus: 1\n    offload:\n      dit: true\n")
     cmd = [
         sys.executable,
         "-m",
-        "fastvideo.entrypoints.openai.api_server",
-        "--model-path",
-        MODEL_PATH,
-        "--host",
+        "fastvideo.entrypoints.cli.main",
+        "serve",
+        "--config",
+        config_path,
+        "--server.host",
         "127.0.0.1",
-        "--port",
+        "--server.port",
         str(port),
-        "--output-dir",
+        "--server.output_dir",
         output_dir,
-        "--num-gpus",
-        "1",
-        "--dit-cpu-offload",
     ]
 
     proc = subprocess.Popen(