Update app.py
Browse files
app.py
CHANGED
|
@@ -2,23 +2,21 @@ import os
|
|
| 2 |
import gradio as gr
|
| 3 |
import torch
|
| 4 |
import spaces
|
| 5 |
-
from gradio import ChatMessage # Importar el objeto correcto para Gradio 6
|
| 6 |
|
| 7 |
# Configuración del token
|
| 8 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
| 9 |
print(f"Token del entorno encontrado: {'Sí' if HF_TOKEN else 'No'}")
|
| 10 |
|
|
|
|
|
|
|
|
|
|
| 11 |
def chat_response(message, history, system_message, max_tokens, temperature, top_p, user_token):
|
| 12 |
"""
|
| 13 |
Genera la respuesta del chat compatible con Gradio 6.0.
|
| 14 |
-
history ahora es una lista de diccionarios o ChatMessage objects: [{'role': 'user', 'content': '...'}, ...]
|
| 15 |
"""
|
| 16 |
-
# Determinar el token a usar
|
| 17 |
token = user_token if user_token else HF_TOKEN
|
| 18 |
|
| 19 |
if not token:
|
| 20 |
-
# Si no hay token, agregamos un mensaje de error al historial
|
| 21 |
-
# Formato correcto: {'role': 'assistant', 'content': '...'}
|
| 22 |
history.append({"role": "assistant", "content": "⚠️ **Error de Configuración**: No se encontró un token de Hugging Face. \n\nConfigura 'HF_TOKEN' en el Space o escribe tu token abajo."})
|
| 23 |
yield history
|
| 24 |
return
|
|
@@ -30,26 +28,24 @@ def chat_response(message, history, system_message, max_tokens, temperature, top
|
|
| 30 |
yield history
|
| 31 |
return
|
| 32 |
|
| 33 |
-
|
|
|
|
| 34 |
|
| 35 |
-
# Construir mensajes
|
| 36 |
-
# history ya viene en el formato correcto para la API en Gradio 6
|
| 37 |
messages = [{"role": "system", "content": system_message}]
|
| 38 |
|
| 39 |
-
# Agregar el historial (ya es una lista de dicts con role/content)
|
| 40 |
for msg in history:
|
| 41 |
-
# Asegurarse de que el mensaje tenga role y content
|
| 42 |
if isinstance(msg, dict) and 'role' in msg and 'content' in msg:
|
| 43 |
messages.append(msg)
|
| 44 |
-
elif hasattr(msg, 'role') and hasattr(msg, 'content'):
|
| 45 |
messages.append({"role": msg.role, "content": msg.content})
|
| 46 |
|
| 47 |
-
# Agregar el mensaje actual del usuario
|
| 48 |
messages.append({"role": "user", "content": message})
|
| 49 |
|
| 50 |
response_text = ""
|
| 51 |
|
| 52 |
try:
|
|
|
|
| 53 |
for chunk in client.chat_completion(
|
| 54 |
messages,
|
| 55 |
max_tokens=int(max_tokens),
|
|
@@ -60,30 +56,15 @@ def chat_response(message, history, system_message, max_tokens, temperature, top
|
|
| 60 |
if chunk.choices and hasattr(chunk.choices[0].delta, 'content') and chunk.choices[0].delta.content:
|
| 61 |
response_text += chunk.choices[0].delta.content
|
| 62 |
|
| 63 |
-
# En Gradio 6, para hacer streaming, debemos actualizar el último mensaje del historial.
|
| 64 |
-
# El historial 'history' contiene los mensajes previos. El último mensaje de usuario está al final.
|
| 65 |
-
# Pero necesitamos agregar el mensaje del asistente que se va construyendo.
|
| 66 |
-
|
| 67 |
-
# Estrategia:
|
| 68 |
-
# 1. Copiar el historial actual (que tiene el mensaje del usuario al final).
|
| 69 |
-
# 2. Si el último elemento es del usuario, lo mantenemos.
|
| 70 |
-
# 3. Añadimos el mensaje del asistente en construcción como un nuevo elemento.
|
| 71 |
-
|
| 72 |
current_history = list(history)
|
| 73 |
-
|
| 74 |
-
# Verificar si el último mensaje es del usuario (lo cual debería serlo al iniciar la respuesta)
|
| 75 |
if current_history and current_history[-1].get('role') == 'user':
|
| 76 |
-
# Añadimos el mensaje del asistente en construcción
|
| 77 |
current_history.append({"role": "assistant", "content": response_text})
|
| 78 |
else:
|
| 79 |
-
# Si por alguna razón el último no es usuario, actualizamos el último asistente
|
| 80 |
-
# (Esto es un caso borde, pero lo manejamos)
|
| 81 |
for i in range(len(current_history) - 1, -1, -1):
|
| 82 |
if current_history[i].get('role') == 'assistant':
|
| 83 |
current_history[i]['content'] = response_text
|
| 84 |
break
|
| 85 |
else:
|
| 86 |
-
# Si no hay asistente, añadimos
|
| 87 |
current_history.append({"role": "assistant", "content": response_text})
|
| 88 |
|
| 89 |
yield current_history
|
|
@@ -95,15 +76,15 @@ def chat_response(message, history, system_message, max_tokens, temperature, top
|
|
| 95 |
error_text = "⚠️ **Token Inválido**: El token de Hugging Face no es válido."
|
| 96 |
elif "429" in error_msg:
|
| 97 |
error_text = "⚠️ **Demasiadas peticiones**: Intenta de nuevo en unos segundos."
|
|
|
|
|
|
|
| 98 |
else:
|
| 99 |
error_text = f"⚠️ **Error**: {error_msg[:200]}"
|
| 100 |
|
| 101 |
current_history = list(history)
|
| 102 |
-
# Actualizar o añadir mensaje de error
|
| 103 |
if current_history and current_history[-1].get('role') == 'user':
|
| 104 |
current_history.append({"role": "assistant", "content": error_text})
|
| 105 |
else:
|
| 106 |
-
# Si no, actualizamos el último asistente existente
|
| 107 |
for i in range(len(current_history) - 1, -1, -1):
|
| 108 |
if current_history[i].get('role') == 'assistant':
|
| 109 |
current_history[i]['content'] = error_text
|
|
@@ -119,10 +100,10 @@ def greet_gpu(n):
|
|
| 119 |
zero = torch.Tensor([0]).to(device)
|
| 120 |
return f"Hello {int(zero + n)} Tensor (GPU: {device})"
|
| 121 |
|
| 122 |
-
# --- INTERFAZ
|
| 123 |
with gr.Blocks() as demo:
|
| 124 |
-
gr.Markdown("# Esme
|
| 125 |
-
gr.Markdown("
|
| 126 |
|
| 127 |
with gr.Row():
|
| 128 |
token_input = gr.Textbox(
|
|
@@ -133,7 +114,6 @@ with gr.Blocks() as demo:
|
|
| 133 |
)
|
| 134 |
clear_btn = gr.Button("🗑️ Limpiar Chat", scale=1)
|
| 135 |
|
| 136 |
-
# Chatbot en Gradio 6 maneja automáticamente el formato de diccionarios
|
| 137 |
chatbot = gr.Chatbot(
|
| 138 |
label="Chat con Esme",
|
| 139 |
height=600,
|
|
@@ -143,7 +123,7 @@ with gr.Blocks() as demo:
|
|
| 143 |
with gr.Row():
|
| 144 |
with gr.Column(scale=3):
|
| 145 |
system_input = gr.Textbox(
|
| 146 |
-
value="
|
| 147 |
label="System Message"
|
| 148 |
)
|
| 149 |
max_tokens_input = gr.Slider(1, 4096, value=512, step=1, label="Max Tokens")
|
|
@@ -155,7 +135,6 @@ with gr.Blocks() as demo:
|
|
| 155 |
|
| 156 |
chat_input = gr.Textbox(label="Mensaje", placeholder="Escribe aquí...")
|
| 157 |
|
| 158 |
-
# Evento de envío
|
| 159 |
chat_input.submit(
|
| 160 |
fn=chat_response,
|
| 161 |
inputs=[
|
|
@@ -170,9 +149,7 @@ with gr.Blocks() as demo:
|
|
| 170 |
outputs=[chatbot]
|
| 171 |
)
|
| 172 |
|
| 173 |
-
|
| 174 |
-
clear_btn.click(lambda: [], None, chatbot) # Devolver lista vacía para limpiar
|
| 175 |
|
| 176 |
if __name__ == "__main__":
|
| 177 |
-
# Tema en launch()
|
| 178 |
demo.queue(max_size=50).launch(theme=gr.themes.Soft())
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import torch
|
| 4 |
import spaces
|
|
|
|
| 5 |
|
| 6 |
# Configuración del token
|
| 7 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
| 8 |
print(f"Token del entorno encontrado: {'Sí' if HF_TOKEN else 'No'}")
|
| 9 |
|
| 10 |
+
# CAMBIO: Usamos un modelo disponible en la API gratuita
|
| 11 |
+
MODEL_ID = "Qwen/Qwen2.5-Coder-7B-Instruct"
|
| 12 |
+
|
| 13 |
def chat_response(message, history, system_message, max_tokens, temperature, top_p, user_token):
|
| 14 |
"""
|
| 15 |
Genera la respuesta del chat compatible con Gradio 6.0.
|
|
|
|
| 16 |
"""
|
|
|
|
| 17 |
token = user_token if user_token else HF_TOKEN
|
| 18 |
|
| 19 |
if not token:
|
|
|
|
|
|
|
| 20 |
history.append({"role": "assistant", "content": "⚠️ **Error de Configuración**: No se encontró un token de Hugging Face. \n\nConfigura 'HF_TOKEN' en el Space o escribe tu token abajo."})
|
| 21 |
yield history
|
| 22 |
return
|
|
|
|
| 28 |
yield history
|
| 29 |
return
|
| 30 |
|
| 31 |
+
# Inicializar cliente con el NUEVO modelo
|
| 32 |
+
client = InferenceClient(token=token, model=MODEL_ID)
|
| 33 |
|
| 34 |
+
# Construir mensajes
|
|
|
|
| 35 |
messages = [{"role": "system", "content": system_message}]
|
| 36 |
|
|
|
|
| 37 |
for msg in history:
|
|
|
|
| 38 |
if isinstance(msg, dict) and 'role' in msg and 'content' in msg:
|
| 39 |
messages.append(msg)
|
| 40 |
+
elif hasattr(msg, 'role') and hasattr(msg, 'content'):
|
| 41 |
messages.append({"role": msg.role, "content": msg.content})
|
| 42 |
|
|
|
|
| 43 |
messages.append({"role": "user", "content": message})
|
| 44 |
|
| 45 |
response_text = ""
|
| 46 |
|
| 47 |
try:
|
| 48 |
+
# Llamada a la API
|
| 49 |
for chunk in client.chat_completion(
|
| 50 |
messages,
|
| 51 |
max_tokens=int(max_tokens),
|
|
|
|
| 56 |
if chunk.choices and hasattr(chunk.choices[0].delta, 'content') and chunk.choices[0].delta.content:
|
| 57 |
response_text += chunk.choices[0].delta.content
|
| 58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
current_history = list(history)
|
|
|
|
|
|
|
| 60 |
if current_history and current_history[-1].get('role') == 'user':
|
|
|
|
| 61 |
current_history.append({"role": "assistant", "content": response_text})
|
| 62 |
else:
|
|
|
|
|
|
|
| 63 |
for i in range(len(current_history) - 1, -1, -1):
|
| 64 |
if current_history[i].get('role') == 'assistant':
|
| 65 |
current_history[i]['content'] = response_text
|
| 66 |
break
|
| 67 |
else:
|
|
|
|
| 68 |
current_history.append({"role": "assistant", "content": response_text})
|
| 69 |
|
| 70 |
yield current_history
|
|
|
|
| 76 |
error_text = "⚠️ **Token Inválido**: El token de Hugging Face no es válido."
|
| 77 |
elif "429" in error_msg:
|
| 78 |
error_text = "⚠️ **Demasiadas peticiones**: Intenta de nuevo en unos segundos."
|
| 79 |
+
elif "not supported" in error_msg.lower():
|
| 80 |
+
error_text = f"⚠️ **Modelo no soportado**: El modelo '{MODEL_ID}' no está disponible en la API gratuita actual. Intenta con otro modelo."
|
| 81 |
else:
|
| 82 |
error_text = f"⚠️ **Error**: {error_msg[:200]}"
|
| 83 |
|
| 84 |
current_history = list(history)
|
|
|
|
| 85 |
if current_history and current_history[-1].get('role') == 'user':
|
| 86 |
current_history.append({"role": "assistant", "content": error_text})
|
| 87 |
else:
|
|
|
|
| 88 |
for i in range(len(current_history) - 1, -1, -1):
|
| 89 |
if current_history[i].get('role') == 'assistant':
|
| 90 |
current_history[i]['content'] = error_text
|
|
|
|
| 100 |
zero = torch.Tensor([0]).to(device)
|
| 101 |
return f"Hello {int(zero + n)} Tensor (GPU: {device})"
|
| 102 |
|
| 103 |
+
# --- INTERFAZ ---
|
| 104 |
with gr.Blocks() as demo:
|
| 105 |
+
gr.Markdown("# Esme ")
|
| 106 |
+
gr.Markdown(f" Modelo actual: **{MODEL_ID}** | 🔐 Token opcional abajo")
|
| 107 |
|
| 108 |
with gr.Row():
|
| 109 |
token_input = gr.Textbox(
|
|
|
|
| 114 |
)
|
| 115 |
clear_btn = gr.Button("🗑️ Limpiar Chat", scale=1)
|
| 116 |
|
|
|
|
| 117 |
chatbot = gr.Chatbot(
|
| 118 |
label="Chat con Esme",
|
| 119 |
height=600,
|
|
|
|
| 123 |
with gr.Row():
|
| 124 |
with gr.Column(scale=3):
|
| 125 |
system_input = gr.Textbox(
|
| 126 |
+
value="Eres un asistente de programación experto y amable llamado Esme.",
|
| 127 |
label="System Message"
|
| 128 |
)
|
| 129 |
max_tokens_input = gr.Slider(1, 4096, value=512, step=1, label="Max Tokens")
|
|
|
|
| 135 |
|
| 136 |
chat_input = gr.Textbox(label="Mensaje", placeholder="Escribe aquí...")
|
| 137 |
|
|
|
|
| 138 |
chat_input.submit(
|
| 139 |
fn=chat_response,
|
| 140 |
inputs=[
|
|
|
|
| 149 |
outputs=[chatbot]
|
| 150 |
)
|
| 151 |
|
| 152 |
+
clear_btn.click(lambda: [], None, chatbot)
|
|
|
|
| 153 |
|
| 154 |
if __name__ == "__main__":
|
|
|
|
| 155 |
demo.queue(max_size=50).launch(theme=gr.themes.Soft())
|