Upload UltravoxPipeline

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/Users/zhuang/repos/ultravox-omni/artifacts/model-zhuang.2025-01-08-v0_5.llama3_1-8b-3d.20345bc:v6",
+  "_name_or_path": "/Users/zhuang/repos/ultravox-omni/artifacts/model-zhuang.2025-01-08-v0_5.llama3_1-8b-3d.20345bc:v5",
   "architectures": [
     "UltravoxModel"
   ],

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03a1d38de8d3e6ab5423e4fba58d97acf3224dfde4227d1ad0243354b5322165
+oid sha256:7e8dd6f0a3e8b091710cbbd92649b5f6bebffe9f1e4f99c2dceb2955cf483545
 size 1374682344

ultravox_pipeline.py

@@ -33,14 +33,15 @@ class UltravoxPipeline(transformers.Pipeline):
                 model.config.audio_model_id or model.config.audio_config._name_or_path
             )
 
-        super().__init__(model=model, tokenizer=tokenizer, **kwargs)
-
         self.processor = UltravoxProcessor(
             audio_processor=audio_processor,
             tokenizer=tokenizer,
             stack_factor=model.config.stack_factor,
+            audio_context_size=model.audio_tower_context_length,
         )
 
+        super().__init__(model=model, tokenizer=tokenizer, **kwargs)
+
     def _sanitize_parameters(self, **kwargs):
         generation_keys = ["temperature", "max_new_tokens", "repetition_penalty"]
         generation_kwargs = {k: kwargs[k] for k in kwargs if k in generation_keys}