fix: preserve Omni processor conditioning on vLLM 0.31

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>

vllm_omni/model_executor/models/cosyvoice3/cosyvoice3.py

@@ -42,7 +42,7 @@ from vllm.v1.sample.ops.topk_topp_sampler import random_sample
 from vllm.v1.sample.sampler import Sampler
 
 from vllm_omni.data_entry_keys import EmbeddingsStruct, OmniPayloadStruct, to_dict, to_struct
-from vllm_omni.inputs.mm_processor import OmniMultiModalProcessor
+from vllm_omni.inputs.mm_processor import OmniDummyInputsBuilder, OmniMultiModalProcessor
 from vllm_omni.model_executor.models.cosyvoice3.ras_sampler import MAX_FUSED_TOP_K, fused_ras_sample
 from vllm_omni.model_executor.models.cosyvoice3.runtime import (
     cosyvoice3_batch_flow_debug,
@@ -675,7 +675,7 @@ class CosyVoice3MultiModalProcessor(OmniMultiModalProcessor[CosyVoice3MultiModal
         ]
 
 
-class CosyVoice3DummyInputsBuilder(BaseDummyInputsBuilder[CosyVoice3MultiModalProcessingInfo]):
+class CosyVoice3DummyInputsBuilder(OmniDummyInputsBuilder[CosyVoice3MultiModalProcessingInfo]):
     def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str:
         return "Hello, this is a test of the CosyVoice3 system capability."
 

vllm_omni/model_executor/models/glm_tts/glm_tts.py

@@ -50,7 +50,7 @@ from vllm.v1.outputs import SamplerOutput
 from vllm.v1.sample.metadata import SamplingMetadata
 from vllm.v1.sample.sampler import Sampler
 
-from vllm_omni.inputs.mm_processor import OmniMultiModalProcessor
+from vllm_omni.inputs.mm_processor import OmniDummyInputsBuilder, OmniMultiModalProcessor
 from vllm_omni.model_executor.models.common.nucleus_ras_sampling import ras_sample_one as _ras_sample_one
 from vllm_omni.model_executor.models.output_templates import OmniOutput
 from vllm_omni.platforms import current_omni_platform
@@ -674,7 +674,7 @@ class GLMTTSMultiModalProcessor(OmniMultiModalProcessor[GLMTTSMultiModalProcessi
         ]
 
 
-class GLMTTSDummyInputsBuilder(BaseDummyInputsBuilder[GLMTTSMultiModalProcessingInfo]):
+class GLMTTSDummyInputsBuilder(OmniDummyInputsBuilder[GLMTTSMultiModalProcessingInfo]):
     def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str:
         return "This is a test of the GLM-TTS voice cloning system."
 

vllm_omni/model_executor/models/mimo_audio/mimo_audio.py

@@ -44,7 +44,7 @@ from vllm.utils.platform_utils import is_pin_memory_available
 from vllm.v1.sample.metadata import SamplingMetadata
 from vllm.v1.sample.sampler import Sampler
 
-from vllm_omni.inputs.mm_processor import OmniMultiModalProcessor
+from vllm_omni.inputs.mm_processor import OmniDummyInputsBuilder, OmniMultiModalProcessor
 from vllm_omni.model_executor.custom_process_mixin import CustomProcessMixin
 from vllm_omni.model_executor.models.mimo_audio.config_mimo_audio import (
     NO_INTERLEAVE_NEXT_TOKEN_ID,
@@ -224,7 +224,7 @@ class MiMoAudioLLMProcessingInfo(
         return {"audio": 1}
 
 
-class MiMoAudioLLMDummyInputsBuilder(BaseDummyInputsBuilder[MiMoAudioLLMProcessingInfo]):
+class MiMoAudioLLMDummyInputsBuilder(OmniDummyInputsBuilder[MiMoAudioLLMProcessingInfo]):
     _processor_inputs_cache: LRUCache = LRUCache(capacity=1024)
 
     def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str:

vllm_omni/model_executor/models/omnivoice/omnivoice.py

@@ -37,7 +37,7 @@ from vllm.multimodal.processing import (
 )
 from vllm.sequence import IntermediateTensors
 
-from vllm_omni.inputs.mm_processor import OmniMultiModalProcessor
+from vllm_omni.inputs.mm_processor import OmniDummyInputsBuilder, OmniMultiModalProcessor
 from vllm_omni.model_executor.models.output_templates import OmniOutput
 from vllm_omni.platforms import current_omni_platform
 from vllm_omni.transformers_utils.configs.omnivoice import OmniVoiceConfig
@@ -241,7 +241,7 @@ class OmniVoiceMultiModalProcessor(OmniMultiModalProcessor[OmniVoiceMultiModalPr
         ]
 
 
-class OmniVoiceDummyInputsBuilder(BaseDummyInputsBuilder[OmniVoiceMultiModalProcessingInfo]):
+class OmniVoiceDummyInputsBuilder(OmniDummyInputsBuilder[OmniVoiceMultiModalProcessingInfo]):
     def get_dummy_text(self, mm_counts: Mapping[str, int]) -> str:
         return "Hello, this is a test of the OmniVoice system."
 

vllm_omni/model_executor/models/voxtral_tts/voxtral_tts_audio_generation.py

@@ -838,6 +838,14 @@ class VoxtralTTSDummyInputsBuilder(BaseDummyInputsBuilder[VoxtralTTSProcessingIn
 
 
 class VoxtralTTSMultiModalProcessor(BaseMultiModalProcessor[VoxtralTTSProcessingInfo]):
+    def get_dummy_inputs(
+        self,
+        seq_len: int,
+        mm_counts: Mapping[str, int],
+        mm_options: Mapping[str, Any],
+    ) -> ProcessorInputs:
+        return self.dummy_inputs.get_dummy_processor_inputs(seq_len, mm_counts, mm_options)
+
     def _get_mm_fields_config(
         self,
         hf_inputs: Mapping[str, NestedTensors],