Update app.py

app.py

@@ -21,17 +21,15 @@ DATA:
 - CONTACT: projects@softstream.tech.
 """
 
-# --- 2. SETUP (SECURE VERSION) ---
+# --- 2. SETUP ---
 hf_token = os.getenv("HF_TOKEN")
 
 if not hf_token:
-    # Fallback explanation if token is missing
     raise ValueError("Error: HF_TOKEN is missing. Go to Settings > Secrets and add it!")
 
 client = InferenceClient(api_key=hf_token)
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
-# Using "tiny" is safest for free CPU tiers
 whisper_model = whisper.load_model("tiny", device=device)
 
 def voice_chat(audio, history):
@@ -48,9 +46,9 @@ def voice_chat(audio, history):
         # B. Think
         history.append({"role": "user", "content": transcription + " (Answer in 1 sentence max)"})
         
-        # --- FIX: SWITCHED MODEL TO PHI-3.5 ---
+        # --- FIX: Using Qwen 2.5 (Most reliable free model right now) ---
         response = client.chat.completions.create(
-            model="microsoft/Phi-3.5-mini-instruct", 
+            model="Qwen/Qwen2.5-7B-Instruct", 
             messages=history, 
             max_tokens=50, 
             temperature=0.3
@@ -67,7 +65,6 @@ def voice_chat(audio, history):
         return audio_path, ai_text, history
 
     except Exception as e:
-        # If Phi-3 also fails, we return the error to the text box to debug
         return None, f"Error: {str(e)}", history
 
 # --- 3. UI ---
@@ -80,7 +77,7 @@ with gr.Blocks(theme="soft") as demo:
         input_audio = gr.Audio(sources=["microphone"], type="filepath", label="Ask Question")
     
     with gr.Row():
-        output_audio = gr.Audio(label="AI Answer (<10s)", autoplay=True)
+        output_audio = gr.Audio(label="AI Answer", autoplay=True)
         output_text = gr.Textbox(label="Transcript")
     
     clear_btn = gr.Button("Reset")