Rename app (22).py to app.py

app (22).py

@@ -1,50 +0,0 @@
-import gradio as gr
-from transformers import pipeline
-import os
-import torch
-
-# --- Performance Improvement ---
-# Configure PyTorch for CPU performance
-num_cpu_cores = os.cpu_count() or 1 # Default to 1 if os.cpu_count() is None
-torch.set_num_threads(num_cpu_cores)
-print(f"✅ PyTorch is configured to use {num_cpu_cores} CPU cores.")
-
-
-# --- Model and Pipeline ---
-# Initialize the pipeline. It will default to the CPU.
-# Using a specific revision for reproducibility
-pipe = pipeline(
-    "audio-classification",
-    model="MIT/ast-finetuned-audioset-10-10-0.4593"
-)
-
-
-# --- Core Logic Function ---
-def classify_audio(audio):
-    """
-    Classifies the audio, takes the top 3 predictions,
-    and formats them into a single, human-readable string.
-    """
-    if audio is None:
-        return "Please upload an audio file first."
-
-    result = pipe(audio)
-    return {label['label']: label['score'] for label in result}
-
-# --- Gradio Interface ---
-# Create the Gradio app interface
-app = gr.Interface(
-    fn=classify_audio,
-    inputs=gr.Audio(type="filepath", label="Upload Audio File"),
-    outputs=gr.Label(num_top_classes=3), # This will now receive a simple string
-    title="Audio Classification with MIT/AST",
-    description=(
-        "Upload an audio file to classify it. The model will identify the top 3 most likely sound categories. "
-    ),
-    cache_examples=False,
-)
-
-# --- App Launch ---
-# Launch the app with sharing enabled for Hugging Face Spaces
-if __name__ == "__main__":
-    app.launch(share=True)

app.py

@@ -0,0 +1,56 @@
+import os
+import torch
+import gradio as gr
+import spaces
+from transformers import pipeline
+
+# ── Аптымізацыя CPU (падае запасны варыянт, калі GPU няма) ──
+num_cpu_cores = os.cpu_count() or 1
+torch.set_num_threads(num_cpu_cores)
+print(f"✅ PyTorch настроены на {num_cpu_cores} ядраў CPU.")
+
+# ── Ініцыялізацыя мадэлі ──
+pipe = pipeline(
+    task="audio-classification",
+    model="MIT/ast-finetuned-audioset-10-10-0.448"
+)
+
+# Флаг, каб не пераносіць мадэль на GPU паўторна
+_model_on_gpu = False
+
+# ── Асноўная функцыя, якая патрабуе GPU ──
+@spaces.GPU(duration=20)       # 90 с хопіць для большасці запытаў
+def classify_audio(audio_path: str):
+    """
+    Класыфікуе аўдыя, вяртае 3 лепшыя тэгі і іх верагоднасці.
+    ZeroGPU выдзяляе GPU толькі на час працы гэтай функцыі.
+    """
+    global _model_on_gpu
+
+    if audio_path is None:
+        return {"⚠️": "Загрузіце аўдыя-файл."}
+
+    # Аднойчы пераносім мадэль на GPU
+    if torch.cuda.is_available() and not _model_on_gpu:
+        pipe.model.to("cuda")
+        _model_on_gpu = True
+
+    preds = pipe(audio_path)         # інферэнс
+    # вяртаем dict label→score для gr.Label
+    return {p["label"]: p["score"] for p in preds[:3]}
+
+# ── Інтэрфейс Gradio ──
+app = gr.Interface(
+    fn=classify_audio,
+    inputs=gr.Audio(type="filepath", label="Upload Audio File"),
+    outputs=gr.Label(num_top_classes=3),
+    title="Audio Classification (MIT/AST) · ZeroGPU",
+    description=(
+        "Загрузіце аўдыя-файл – мадэль дасць тры найбольш верагодныя катэгорыі гуку."
+    ),
+    cache_examples=False,
+)
+
+# ── Запуск (у Spaces прапускаем share=True – яго непатрэбна) ──
+if __name__ == "__main__":
+    app.launch()