Update app.py

app.py

@@ -1,38 +1,16 @@
-import os
-
-# Fix PyTorch Inductor cache directory and HF cache permission issues on Hugging Face Spaces
-os.environ["TORCHINDUCTOR_CACHE_DIR"] = "/tmp"
-os.environ["HF_HOME"] = "/tmp"
-
 from fastapi import FastAPI
 from pydantic import BaseModel
-from transformers import AutoProcessor
-from transformers.models.vits.modeling_vits import VitsForConditionalGeneration
-import torch
-import io
-from fastapi.responses import StreamingResponse
-import soundfile as sf
+from transformers import pipeline
 
 app = FastAPI()
 
-model_name = "Somali-tts/somali_tts_model"
-processor = AutoProcessor.from_pretrained(model_name)
-model = VitsForConditionalGeneration.from_pretrained(model_name)
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-model.to(device)
+# Load your model pipeline once on startup
+summarizer = pipeline("text2text-generation", model="zakihassan04/tacab_ai_beero")
 
-class TextInput(BaseModel):
+class TextRequest(BaseModel):
     inputs: str
 
-@app.post("/synthesize")
-async def synthesize_tts(data: TextInput):
-    inputs = processor(data.inputs, return_tensors="pt").to(device)
-    with torch.no_grad():
-        audio = model.generate(**inputs)
-    audio = audio.squeeze().cpu().numpy()
-
-    buf = io.BytesIO()
-    sf.write(buf, audio, samplerate=22050, format="WAV")
-    buf.seek(0)
-
-    return StreamingResponse(buf, media_type="audio/wav")
+@app.post("/generate")
+async def generate_text(request: TextRequest):
+    out = summarizer(request.inputs, max_length=200, do_sample=False)
+    return {"generated_text": out[0]["generated_text"]}