Update app.py

app.py

@@ -1,40 +1,18 @@
-import streamlit as st
-import base64
-import io
 from huggingface_hub import InferenceClient
-from gtts import gTTS
-from audiorecorder import audiorecorder
-import speech_recognition as sr
-from pydub import AudioSegment
+import gradio as gr
 
-if "history" not in st.session_state:
-    st.session_state.history = []
-
-def recognize_speech(audio_data, show_messages=True):
-    recognizer = sr.Recognizer()
-    audio_recording = sr.AudioFile(audio_data)
-
-    with audio_recording as source:
-        audio = recognizer.record(source)
-
-    try:
-        audio_text = recognizer.recognize_google(audio, language="es-ES")
-        if show_messages:
-            st.subheader("Texto Reconocido:")
-            st.write(audio_text)
-            st.success("Reconocimiento de voz completado.")
-    except sr.UnknownValueError:
-        st.warning("No se pudo reconocer el audio. ¿Intentaste grabar algo?")
-        audio_text = ""
-    except sr.RequestError:
-        st.error("Hablame para comenzar!")
-        audio_text = ""
-
-    return audio_text
+client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+system_prompt = "Te llamarás Caman 2.0 y tus respuestas serán breves"
+system_prompt_sent = False
 
 def format_prompt(message, history):
+    global system_prompt_sent
     prompt = "<s>"
 
+    if not any(f"[INST] {system_prompt} [/INST]" in user_prompt for user_prompt, _ in history):
+        prompt += f"[INST] {system_prompt} [/INST]"
+        system_prompt_sent = True
+
     for user_prompt, bot_response in history:
         prompt += f"[INST] {user_prompt} [/INST]"
         prompt += f" {bot_response}</s> "
@@ -42,10 +20,11 @@ def format_prompt(message, history):
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
-def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.95, repetition_penalty=1.0):
-    client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
-
-    temperature = float(temperature) if temperature is not None else 0.9
+def generate(
+    prompt, history, temperature=0.9, max_new_tokens=2048, top_p=0.95, repetition_penalty=1.0,
+):
+    global system_prompt_sent
+    temperature = float(temperature)
     if temperature < 1e-2:
         temperature = 1e-2
     top_p = float(top_p)
@@ -59,55 +38,26 @@ def generate(audio_text, history, temperature=None, max_new_tokens=512, top_p=0.
         seed=42,
     )
 
-    formatted_prompt = format_prompt(audio_text, history)
-    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
-    response = ""
-
-    for response_token in stream:
-        response += response_token.token.text
+    formatted_prompt = format_prompt(prompt, history)
     
-    response = ' '.join(response.split()).replace('</s>', '')
-    audio_file = text_to_speech(response, speed=1.3)
-    return response, audio_file
-
-def text_to_speech(text, speed=1.3):
-    tts = gTTS(text=text, lang='es')
-    audio_fp = io.BytesIO()
-    tts.write_to_fp(audio_fp)
-    audio_fp.seek(0)
-    audio = AudioSegment.from_file(audio_fp, format="mp3")
-    modified_speed_audio = audio.speedup(playback_speed=speed)
-    modified_audio_fp = io.BytesIO()
-    modified_speed_audio.export(modified_audio_fp, format="mp3")
-    modified_audio_fp.seek(0)
-    return modified_audio_fp
-
-def main():
-    audio_data = audiorecorder("Habla para grabar", "Deteniendo la grabación...")
-
-    if not audio_data.empty():
-        st.audio(audio_data.export().read(), format="audio/wav")
-        audio_data.export("audio.wav", format="wav")
-        audio_text = recognize_speech("audio.wav")
-
-        if not st.session_state.history:
-            pre_prompt = "Te Llamarás Chaman 4.0 y tus respuestas serán sumamente breves."
-            output, _ = generate(pre_prompt, history=st.session_state.history)
-            st.session_state.history.append((pre_prompt, output))
-        
-        if audio_text:
-            output, audio_file = generate(audio_text, history=st.session_state.history)  
-
-            if audio_text:  
-                st.session_state.history.append((audio_text, output))  
-
-            if audio_file is not None:
-                st.markdown(
-                    f"""
-                    <audio autoplay="autoplay" controls="controls" src="data:audio/mp3;base64,{base64.b64encode(audio_file.read()).decode()}" type="audio/mp3" id="audio_player"></audio>
-                    """,
-                    unsafe_allow_html=True
-                )
-
-if __name__ == "__main__":
-    main()
+    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
+    output = ""
+
+    for response in stream:
+        output += response.token.text
+        yield output
+
+    return output
+
+chat_interface = gr.ChatInterface(
+    fn=generate,
+    chatbot=gr.Chatbot(show_label=False, show_share_button=False, show_copy_button=False, likeable=False, layout="vertical", height=700),
+    concurrency_limit=9,
+    theme="soft",
+    retry_btn=None,
+    undo_btn=None,
+    clear_btn=None,
+    submit_btn="Enviar",
+)
+
+chat_interface.launch(show_api=False)