[New Model] support nemotron Audex (#4976)

Signed-off-by: Yuekai Zhang <zhangyuekai@foxmail.com>
Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Yueqian Lin <linyueqian@outlook.com>
Co-authored-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>

tests/dfx/reliability/invalid_param_test/test_invalid_audio_speech.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""HTTP + WebSocket validation for Qwen3-TTS Base and Higgs Audio V3 ``/v1/audio/speech`` (and related routes)."""
+"""HTTP + WebSocket validation for Qwen3-TTS Base, Higgs Audio V3 and Audex TTA ``/v1/audio/speech`` (and related routes)."""
 
 from __future__ import annotations
 
@@ -70,6 +70,19 @@ _HIGGS_AUDIO_V3_SPEECH = [
     ),
 ]
 
+_AUDEX_TTA_CAPTION = "Heavy rain falling on a tin roof."
+_AUDEX_TTA_SPEECH = [
+    pytest.param(
+        OmniServerParams(
+            model="nvidia/Nemotron-Labs-Audex-2B",
+            stage_config_path=get_deploy_config_path("audex_tta.yaml"),
+            server_args=_SPEECH_SERVER_ARGS,
+        ),
+        id="audex_tta",
+        marks=_L4_SPEECH_HW,
+    ),
+]
+
 
 def _apply_batch_overrides(body: dict[str, object], *, loc: str, overrides: dict[str, object]) -> None:
     if loc == "batch":
@@ -325,6 +338,37 @@ def test_speech_higgs_invalid_field_values(
     )
 
 
+# Audex TTA (``audex_tta`` deployment): the serving layer builds the caption
+# prompt and injects the RVQ phase contract + default guidance server-side, so
+# guidance knobs and speech-only fields must be rejected before generation.
+@pytest.mark.parametrize(
+    "overrides, err_message",
+    [
+        pytest.param({"extra_params": {"cfg_scale": 99}}, "cfg_scale", id="cfg_scale_out_of_range"),
+        pytest.param({"extra_params": {"cfg_scale": "big"}}, "cfg_scale", id="cfg_scale_wrong_type"),
+        pytest.param({"extra_params": {"tta_rvq": {}}}, "tta_rvq", id="tta_rvq_not_overridable"),
+        pytest.param({"input": "   "}, "caption", id="caption_whitespace_only"),
+        pytest.param({"voice": "alloy"}, "voice", id="voice_not_supported"),
+        pytest.param(
+            {"ref_audio": "https://example.com/ref.wav"},
+            "reference audio",
+            id="ref_audio_not_supported",
+        ),
+    ],
+)
+@pytest.mark.parametrize("omni_server", _AUDEX_TTA_SPEECH, indirect=True)
+def test_speech_audex_tta_invalid_field_values(
+    omni_server: OmniServer,
+    openai_client: OpenAIClientHandler,
+    overrides: dict[str, object],
+    err_message: str | tuple[str, ...],
+) -> None:
+    body = {"model": omni_server.model, "input": _AUDEX_TTA_CAPTION, **overrides}
+    openai_client.send_audio_speech_http_request(
+        {"json": body, "timeout": 120, "err_code": 400, "err_message": err_message}
+    )
+
+
 # ─────────────────────────────────────────────────────────────────────────────
 # POST /v1/audio/speech/batch (JSON)
 # ─────────────────────────────────────────────────────────────────────────────