Cache pipeline by model+device only, not call-time params (#499)

top_k, function_to_apply, and batch_size are call-time parameters
passed at pipe() invocation, not pipeline construction config.
Including them in the cache key created duplicate pipeline instances
in memory for each unique parameter combination.

Now a single pipeline is reused regardless of top_k/function_to_apply
values — these are passed per-call at inference time.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: satra

src/senselab/audio/tasks/classification/huggingface.py

@@ -44,16 +44,16 @@ def _get_hf_audio_classification_pipeline(
         device, _ = _select_device_and_dtype(
             user_preference=device, compatible_devices=[DeviceType.CUDA, DeviceType.CPU]
         )
-        key = f"{model.path_or_uri}-{model.revision}-{top_k}-{function_to_apply}-{device.value}"
+        # Cache by model identity + device only. top_k, function_to_apply,
+        # and batch_size are call-time parameters passed at pipe() invocation.
+        key = f"{model.path_or_uri}-{model.revision}-{device.value}"
         if key not in cls._pipelines:
             cls._pipelines[key] = cast(
                 Pipeline,
                 pipeline(  # type: ignore[call-overload]
                     task="audio-classification",
                     model=model.path_or_uri,
                     revision=model.revision,
-                    top_k=top_k,
-                    function_to_apply=function_to_apply,
                     device=device.value,
                 ),
             )