Update app.py

app.py

@@ -4,15 +4,15 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
 # -------------------------------
-# Load Qwen2.5-7B for text-only chat
+# Load Qwen2.5-7B (text-only model, works for both Genie & Wizard replies)
 # -------------------------------
 model_id = "Qwen/Qwen2.5-7B-Instruct"
 
 tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
     model_id,
-    torch_dtype=torch.float16,     # use half precision for GPU memory
-    device_map="auto",             # map layers automatically to GPU
+    torch_dtype=torch.float16,
+    device_map="auto",
     trust_remote_code=True
 )
 
@@ -24,40 +24,60 @@ qwen_pipe = pipeline(
 )
 
 # -------------------------------
-# Life Genie function
+# Life Genie: text chat
 # -------------------------------
-@spaces.GPU  # ensure HF detects GPU usage
+@spaces.GPU
 def life_genie(user_input: str) -> str:
-    """
-    Life Genie: magical assistant powered by Qwen2.5-7B.
-    """
     if not user_input.strip():
         return "🧞 Please ask me something magical!"
-    
     prompt = f"You are a witty magical genie. Answer creatively.\nUser: {user_input}\nGenie:"
     output = qwen_pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7)
     return output[0]["generated_text"]
 
+# -------------------------------
+# Whisper ASR for voice input
+# -------------------------------
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-tiny", device=-1)
+
+# -------------------------------
+# Voice Wizard: speech → Genie reply
+# -------------------------------
+@spaces.GPU
+def voice_wizard(audio_file) -> str:
+    if audio_file is None:
+        return "🎤 Please record your voice first!"
+    # Step 1: Transcribe microphone audio
+    transcript = asr_pipe(audio_file)["text"]
+    # Step 2: Ask Genie with that transcript
+    prompt = f"The user spoke: '{transcript}'. Respond like a magical wizard with wisdom and humor.\nWizard:"
+    output = qwen_pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7)
+    return f"✨ Voice Transcription: {transcript}\n\n🔮 Wizard Replies:\n{output[0]['generated_text']}"
+
 # -------------------------------
 # Gradio UI
 # -------------------------------
 with gr.Blocks(css="""
 body {
-    background: linear-gradient(135deg, #ff9966, #ff5e62);
+    background: linear-gradient(135deg, #ff9a9e, #fad0c4);
     font-family: Trebuchet MS, sans-serif;
 }
 """) as demo:
 
-    gr.HTML("<h1 style='text-align:center; color:white;'>🧞✨ Life Genie — Ask Your Magical Assistant ✨🧞</h1>")
-    gr.Markdown("Type your wish below and summon the genie for a response!")
+    gr.HTML("<h1 style='text-align:center; color:white;'>🌈✨ Magic-Lab ✨🌈</h1>")
+    gr.Markdown("Welcome! Meet the Genie through text or voice. 🧞🔮")
 
-    with gr.Row():
-        with gr.Column(scale=1):
-            user_box = gr.Textbox(lines=3, placeholder="E.g. Cheer me up about Mondays 🪄", label="Your Wish")
-            submit_btn = gr.Button("✨ Summon Genie")
-        with gr.Column(scale=1):
-            genie_out = gr.Textbox(lines=10, label="Genie Responds")
+    # Life Genie (Text)
+    with gr.Tab("🧞 Life Genie"):
+        txt_in = gr.Textbox(lines=3, placeholder="Ask me anything...", label="Your Wish")
+        txt_out = gr.Textbox(lines=10, label="Genie's Response")
+        btn_chat = gr.Button("✨ Summon Genie")
+        btn_chat.click(life_genie, inputs=txt_in, outputs=txt_out)
 
-    submit_btn.click(life_genie, inputs=user_box, outputs=genie_out)
+    # Voice Wizard (Voice)
+    with gr.Tab("🔮 Voice Wizard"):
+        aud_in = gr.Audio(sources=["microphone"], type="filepath", label="Speak your spell 🎤")
+        aud_out = gr.Textbox(lines=10, label="Wizard's Reply")
+        btn_voice = gr.Button("🎙️ Cast Spell")
+        btn_voice.click(voice_wizard, inputs=aud_in, outputs=aud_out)
 
 demo.launch()