[Model] Refactor Ultravox to use merged input processor (#11198)

Signed-off-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>

tests/distributed/test_pipeline_parallel.py

@@ -214,7 +214,7 @@ MULTIMODAL_MODELS = {
     "Qwen/Qwen-VL-Chat": PPTestSettings.fast(trust_remote_code=True),
     "Qwen/Qwen2-Audio-7B-Instruct": PPTestSettings.fast(),
     "Qwen/Qwen2-VL-2B-Instruct": PPTestSettings.fast(),
-    "fixie-ai/ultravox-v0_3": PPTestSettings.fast(),
+    "fixie-ai/ultravox-v0_3": PPTestSettings.fast(trust_remote_code=True),
     # [Encoder-decoder]
     # TODO: Implement PP
     # "meta-llama/Llama-3.2-11B-Vision-Instruct": PPTestSettings.fast(),

tests/entrypoints/openai/test_audio.py

@@ -25,6 +25,7 @@ def server():
         "--max-num-seqs",
         "5",
         "--enforce-eager",
+        "--trust-remote-code",
     ]
 
     with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:

tests/models/decoder_only/audio_language/test_ultravox.py

@@ -16,7 +16,7 @@ MODEL_NAME = "fixie-ai/ultravox-v0_3"
 
 AudioTuple = Tuple[np.ndarray, int]
 
-VLLM_PLACEHOLDER = "<|reserved_special_token_0|>"
+VLLM_PLACEHOLDER = "<|audio|>"
 HF_PLACEHOLDER = "<|audio|>"
 
 CHUNKED_PREFILL_KWARGS = {
@@ -46,7 +46,8 @@ def audio(request):
 def server(request, audio_assets):
     args = [
         "--dtype=bfloat16", "--max-model-len=4096", "--enforce-eager",
-        f"--limit-mm-per-prompt=audio={len(audio_assets)}"
+        f"--limit-mm-per-prompt=audio={len(audio_assets)}",
+        "--trust-remote-code"
     ] + [
         f"--{key.replace('_','-')}={value}"
         for key, value in request.param.items()