Update app.py

app.py

@@ -22,23 +22,27 @@ tokenizer = AutoTokenizer.from_pretrained(model_id)
 
 model = AutoModelForCausalLM.from_pretrained(
     model_id,
-    torch_dtype=torch.float16,
-    device_map="auto"
+    torch_dtype=torch.float32
 )
 
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model.to(device)
+
 # =========================
 # Clean Text Using Gemma
 # =========================
 def clean_text(text):
 
+    text = text[:1500]
+
     prompt = f"""
-You are an assistant that cleans Arabic speech transcripts.
+Clean this Arabic speech transcript.
+
+Remove filler words like:
+اممم، آآآ، يعني
 
-Tasks:
-- remove filler words (اممم، آآآ، يعني)
-- remove repeated words
-- keep the same meaning
-- return only the cleaned text
+Remove repeated words.
+Keep the same meaning.
 
 Transcript:
 {text}
@@ -47,17 +51,19 @@ Transcript:
     inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
 
     with torch.no_grad():
+
         outputs = model.generate(
             **inputs,
-            max_new_tokens=200,
-            temperature=0.2
+            max_new_tokens=120,
+            do_sample=True,
+            temperature=0.7,
+            top_p=0.9,
+            repetition_penalty=1.2
         )
 
     result = tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-    return result
-
-# =========================
+    return result# =========================
 # ElevenLabs Speech To Text
 # =========================