Fix [ASR Inference Pipelines]: Minor bug fixes

nemo/collections/asr/inference/model_wrappers/asr_inference_wrapper.py

@@ -92,7 +92,7 @@ def load_model(model_name: str, map_location: torch.device) -> ASRModel:
             asr_model.eval()
             return asr_model
         except Exception as e:
-            raise RuntimeError(f"Failed to load model {model_name}: {str(e)}")
+            raise RuntimeError(f"Failed to load model {model_name}: {str(e)}") from e
 
     @property
     def word_separator(self) -> str:

nemo/collections/asr/inference/model_wrappers/salm_asr_inference_wrapper.py

@@ -85,7 +85,7 @@ def load_model(model_name: str, device: torch.device) -> SALM:
             model.to(device)
             return model
         except Exception as e:
-            raise RuntimeError(f"Failed to load model {model_name}: {str(e)}")
+            raise RuntimeError(f"Failed to load model {model_name}: {str(e)}") from e
 
     def get_window_stride(self) -> float:
         """Returns the window stride of the model."""

nemo/collections/asr/inference/nmt/llm_translator.py

@@ -167,7 +167,7 @@ def load_model(self, llm_params: dict) -> LLM:
             model = LLM(model=self.model_name, **llm_params)
             return model
         except Exception as e:
-            raise RuntimeError(f"Model loading failed: {str(e)}")
+            raise RuntimeError(f"Model loading failed: {str(e)}") from e
 
     def translate_batch(
         self,

nemo/collections/asr/inference/streaming/buffering/cache_feature_bufferer.py

@@ -111,6 +111,8 @@ def free_slots(self, slot_ids: list[int]) -> None:
             slot_ids (list[int]): list of slot ids
         """
         for slot_id in slot_ids:
+            if slot_id not in self.slotidx2streamidx:
+                continue
             self.available_slots.put(slot_id)
             stream_id = self.slotidx2streamidx[slot_id]
             del self.slotidx2streamidx[slot_id], self.streamidx2slotidx[stream_id]
@@ -148,7 +150,7 @@ def preprocess(
         right_padding = torch.floor(right_paddings / self.sample_rate / self.timestep_duration)  # B
         return features, right_padding
 
-    def _update_feature_buffer(self, slot_ids: int, feat_chunk: Tensor) -> None:
+    def _update_feature_buffer(self, slot_ids: list[int], feat_chunk: Tensor) -> None:
         """
         Add an extracted feature to `feature_buffer`
         Args:

nemo/collections/asr/inference/streaming/endpointing/greedy/greedy_endpointing.py

@@ -265,7 +265,7 @@ def detect_eou_near_pivot(
             raise ValueError("Pivot point is out of range")
 
         if search_start_point > pivot_point:
-            raise ValueError("Search start point is greater then pivot_point")
+            raise ValueError("Search start point is greater than pivot_point")
 
         if self.residue_tokens_at_end > 0:
             sequence_length = max(0, sequence_length - self.residue_tokens_at_end)

nemo/collections/asr/inference/streaming/framing/request_options.py

@@ -27,14 +27,14 @@ class ASRRequestOptions:
     None value means that the option is not set and the default value will be used
     """
 
-    enable_itn: bool = None
-    enable_pnc: bool = None
-    stop_history_eou: int = None
-    asr_output_granularity: ASROutputGranularity | str = None
+    enable_itn: bool | None = None
+    enable_pnc: bool | None = None
+    stop_history_eou: int | None = None
+    asr_output_granularity: ASROutputGranularity | str | None = None
     language_code: str | None = None
-    enable_nmt: bool = None
-    source_language: str = None
-    target_language: str = None
+    enable_nmt: bool | None = None
+    source_language: str | None = None
+    target_language: str | None = None
     biasing_cfg: BiasingRequestItemConfig | None = None
 
     def __post_init__(self) -> None:

nemo/collections/asr/inference/streaming/state/cache_aware_state.py

@@ -69,6 +69,11 @@ def update_label_buffer(self, labels: list[int]) -> None:
             labels: (list[int]) list of labels
         """
         shift = len(labels)
+        if shift == 0:
+            return
+        if shift >= len(self.label_buffer):
+            self.label_buffer[:] = labels[-len(self.label_buffer) :]
+            return
         self.label_buffer[:-shift] = self.label_buffer[shift:].copy()
         self.label_buffer[-shift:] = labels.copy()
 

nemo/collections/asr/inference/utils/device_utils.py

@@ -48,10 +48,11 @@ def setup_device(device: str, device_id: int | None, compute_dtype: str) -> tupl
             logging.warning(f"Device ID {device_id} is not available. Using GPU 0 instead.")
             device_id = 0
 
-        compute_dtype = COMPUTE_DTYPE_MAP.get(compute_dtype, None)
+        compute_dtype_str = compute_dtype
+        compute_dtype = COMPUTE_DTYPE_MAP.get(compute_dtype_str, None)
         if compute_dtype is None:
             raise ValueError(
-                f"Invalid compute dtype: {compute_dtype}. Must be one of {list(COMPUTE_DTYPE_MAP.keys())}"
+                f"Invalid compute dtype: {compute_dtype_str}. Must be one of {list(COMPUTE_DTYPE_MAP.keys())}"
             )
 
         device_str = f"cuda:{device_id}"