fix: vLLM >=0.21 compat — EmbedsPrompt moved from vllm.inputs.data to vllm.inputs

Added try/except fallback for both old (0.12.x) and new (0.21+) vLLM API paths.
Bump version to 1.3.5.

Author: LauraGPT

funasr/models/fun_asr_nano/inference_vllm.py

@@ -167,7 +167,10 @@ def __init__(
         **kwargs,
     ):
         from vllm import LLM, SamplingParams
-        from vllm.inputs.data import EmbedsPrompt
+        try:
+            from vllm.inputs import EmbedsPrompt
+        except ImportError:
+            from vllm.inputs.data import EmbedsPrompt
 
         self.device = device
         self.dtype = dtype
@@ -519,7 +522,10 @@ def generate(
             List of result dicts: [{"key": str, "text": str, "timestamps": [...]}]
         """
         from vllm import SamplingParams
-        from vllm.inputs.data import EmbedsPrompt
+        try:
+            from vllm.inputs import EmbedsPrompt
+        except ImportError:
+            from vllm.inputs.data import EmbedsPrompt
 
         if isinstance(inputs, (str, np.ndarray, torch.Tensor)):
             inputs = [inputs]

funasr/models/fun_asr_nano/inference_vllm_pipeline.py

@@ -199,7 +199,10 @@ def _process_one(self, audio_path, **kwargs):
         # Step 3: Batch ASR via vLLM
         # Encode all segments and build prompts
         from vllm import SamplingParams
-        from vllm.inputs.data import EmbedsPrompt
+        try:
+            from vllm.inputs import EmbedsPrompt
+        except ImportError:
+            from vllm.inputs.data import EmbedsPrompt
 
         prompts = []
         for seg_audio in segment_audios:

funasr/models/fun_asr_nano/inference_vllm_streaming.py

@@ -210,7 +210,10 @@ def streaming_generate(self, audio_input, chunk_ms=None, rollback_chars=None,
              "is_final": bool, "chunk_idx": int, "audio_duration_ms": float}
         """
         from vllm import SamplingParams
-        from vllm.inputs.data import EmbedsPrompt
+        try:
+            from vllm.inputs import EmbedsPrompt
+        except ImportError:
+            from vllm.inputs.data import EmbedsPrompt
         from funasr.utils.load_utils import load_audio_text_image_video
 
         chunk_ms = chunk_ms or self.chunk_ms

funasr/models/glm_asr/inference_vllm.py

@@ -192,7 +192,10 @@ def generate(self, inputs, prompt="转录以下音频内容", max_new_tokens=500
             List of {"key": str, "text": str}
         """
         from vllm import SamplingParams
-        from vllm.inputs.data import EmbedsPrompt
+        try:
+            from vllm.inputs import EmbedsPrompt
+        except ImportError:
+            from vllm.inputs.data import EmbedsPrompt
 
         if isinstance(inputs, (str, np.ndarray, torch.Tensor)):
             inputs = [inputs]

funasr/version.txt

@@ -1 +1 @@
-1.3.4
+1.3.5