Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,10 +1,13 @@
|
|
1 |
import os
|
2 |
import time
|
|
|
|
|
|
|
3 |
import google.generativeai as genai
|
4 |
import gradio as gr
|
5 |
from dotenv import load_dotenv
|
6 |
|
7 |
-
# Cargar las variables de entorno
|
8 |
load_dotenv()
|
9 |
API_KEY = os.getenv("GOOGLE_API_KEY")
|
10 |
|
@@ -12,16 +15,15 @@ if not API_KEY:
|
|
12 |
raise ValueError("La clave de API 'GOOGLE_API_KEY' no est谩 configurada en el archivo .env")
|
13 |
|
14 |
# Configuraci贸n del modelo Gemini
|
|
|
15 |
generation_config = {
|
16 |
-
"temperature":
|
17 |
-
"top_p": 0.
|
18 |
"top_k": 40,
|
19 |
"max_output_tokens": 8192,
|
20 |
"response_mime_type": "text/plain",
|
21 |
}
|
22 |
|
23 |
-
genai.configure(api_key=API_KEY)
|
24 |
-
|
25 |
model = genai.GenerativeModel(
|
26 |
model_name="gemini-1.5-flash",
|
27 |
generation_config=generation_config,
|
@@ -30,33 +32,94 @@ model = genai.GenerativeModel(
|
|
30 |
# Inicializar la sesi贸n de chat
|
31 |
chat = model.start_chat(history=[])
|
32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
# Funci贸n para transformar el historial de Gradio al formato de Gemini
|
34 |
def transform_history(history):
|
|
|
35 |
new_history = []
|
36 |
for chat in history:
|
37 |
-
|
38 |
-
|
|
|
|
|
39 |
return new_history
|
40 |
|
41 |
-
# Funci贸n
|
42 |
def response(message, history):
|
|
|
43 |
global chat
|
44 |
|
45 |
# Transformar el historial al formato esperado por Gemini
|
46 |
chat.history = transform_history(history)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
-
#
|
49 |
-
|
|
|
50 |
response.resolve()
|
51 |
|
52 |
-
#
|
53 |
for i in range(len(response.text)):
|
54 |
time.sleep(0.01)
|
55 |
yield response.text[: i + 1]
|
56 |
|
57 |
-
#
|
58 |
-
gr.ChatInterface(
|
59 |
response,
|
60 |
-
|
61 |
-
|
62 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
import time
|
3 |
+
import uuid
|
4 |
+
from typing import List, Tuple, Optional, Union
|
5 |
+
from PIL import Image
|
6 |
import google.generativeai as genai
|
7 |
import gradio as gr
|
8 |
from dotenv import load_dotenv
|
9 |
|
10 |
+
# Cargar las variables de entorno desde el archivo .env
|
11 |
load_dotenv()
|
12 |
API_KEY = os.getenv("GOOGLE_API_KEY")
|
13 |
|
|
|
15 |
raise ValueError("La clave de API 'GOOGLE_API_KEY' no est谩 configurada en el archivo .env")
|
16 |
|
17 |
# Configuraci贸n del modelo Gemini
|
18 |
+
genai.configure(api_key=API_KEY)
|
19 |
generation_config = {
|
20 |
+
"temperature": 0.7,
|
21 |
+
"top_p": 0.9,
|
22 |
"top_k": 40,
|
23 |
"max_output_tokens": 8192,
|
24 |
"response_mime_type": "text/plain",
|
25 |
}
|
26 |
|
|
|
|
|
27 |
model = genai.GenerativeModel(
|
28 |
model_name="gemini-1.5-flash",
|
29 |
generation_config=generation_config,
|
|
|
32 |
# Inicializar la sesi贸n de chat
|
33 |
chat = model.start_chat(history=[])
|
34 |
|
35 |
+
# Constantes para el manejo de im谩genes
|
36 |
+
IMAGE_CACHE_DIRECTORY = "/tmp"
|
37 |
+
IMAGE_WIDTH = 512
|
38 |
+
CHAT_HISTORY = List[Tuple[Optional[Union[Tuple[str], str]], Optional[str]]]
|
39 |
+
|
40 |
+
# Funci贸n para preprocesar una imagen
|
41 |
+
def preprocess_image(image: Image.Image) -> Optional[Image.Image]:
|
42 |
+
"""Redimensiona una imagen manteniendo la relaci贸n de aspecto."""
|
43 |
+
if image:
|
44 |
+
image_height = int(image.height * IMAGE_WIDTH / image.width)
|
45 |
+
return image.resize((IMAGE_WIDTH, image_height))
|
46 |
+
|
47 |
+
# Funci贸n para almacenar una imagen en cach茅
|
48 |
+
def cache_pil_image(image: Image.Image) -> str:
|
49 |
+
"""Guarda la imagen como archivo JPEG en un directorio temporal."""
|
50 |
+
image_filename = f"{uuid.uuid4()}.jpeg"
|
51 |
+
os.makedirs(IMAGE_CACHE_DIRECTORY, exist_ok=True)
|
52 |
+
image_path = os.path.join(IMAGE_CACHE_DIRECTORY, image_filename)
|
53 |
+
image.save(image_path, "JPEG")
|
54 |
+
return image_path
|
55 |
+
|
56 |
# Funci贸n para transformar el historial de Gradio al formato de Gemini
|
57 |
def transform_history(history):
|
58 |
+
"""Transforma el historial del formato de Gradio al formato que Gemini espera."""
|
59 |
new_history = []
|
60 |
for chat in history:
|
61 |
+
if chat[0]: # Mensaje del usuario
|
62 |
+
new_history.append({"parts": [{"text": chat[0]}], "role": "user"})
|
63 |
+
if chat[1]: # Respuesta del modelo
|
64 |
+
new_history.append({"parts": [{"text": chat[1]}], "role": "model"})
|
65 |
return new_history
|
66 |
|
67 |
+
# Funci贸n principal para manejar las respuestas del chat
|
68 |
def response(message, history):
|
69 |
+
"""Maneja la interacci贸n multimodal y env铆a texto e im谩genes al modelo."""
|
70 |
global chat
|
71 |
|
72 |
# Transformar el historial al formato esperado por Gemini
|
73 |
chat.history = transform_history(history)
|
74 |
+
|
75 |
+
# Obtener el texto del mensaje y las im谩genes cargadas
|
76 |
+
text_prompt = message["text"]
|
77 |
+
files = message["files"]
|
78 |
+
|
79 |
+
# Procesar im谩genes cargadas
|
80 |
+
image_prompts = [preprocess_image(Image.open(file).convert('RGB')) for file in files] if files else []
|
81 |
+
if files:
|
82 |
+
for file in files:
|
83 |
+
image = Image.open(file).convert('RGB')
|
84 |
+
image_preview = preprocess_image(image)
|
85 |
+
if image_preview:
|
86 |
+
# Guardar la imagen y obtener la ruta
|
87 |
+
image_path = cache_pil_image(image)
|
88 |
+
# Leer la imagen en formato binario para enviarla como Blob
|
89 |
+
with open(image_path, "rb") as img_file:
|
90 |
+
img_data = img_file.read()
|
91 |
+
# Crear un diccionario con los datos binarios y su tipo MIME
|
92 |
+
image_prompt = {
|
93 |
+
"mime_type": "image/jpeg",
|
94 |
+
"data": img_data
|
95 |
+
}
|
96 |
+
image_prompts.append(image_prompt)
|
97 |
+
|
98 |
+
# Aqu铆 podemos retornar el listado de `image_prompts` o las rutas
|
99 |
+
return image_prompts # Esto puede ser 煤til si luego quieres usar estos datos en otro lugar
|
100 |
|
101 |
+
# Combinar texto e im谩genes para el modelo
|
102 |
+
prompts = [text_prompt] + image_prompts
|
103 |
+
response = chat.send_message(prompts)
|
104 |
response.resolve()
|
105 |
|
106 |
+
# Generar respuesta car谩cter por car谩cter para una experiencia m谩s fluida
|
107 |
for i in range(len(response.text)):
|
108 |
time.sleep(0.01)
|
109 |
yield response.text[: i + 1]
|
110 |
|
111 |
+
# Crear la interfaz de usuario
|
112 |
+
demo = gr.ChatInterface(
|
113 |
response,
|
114 |
+
examples=[{"text": "Describe the image:", "files": []}],
|
115 |
+
multimodal=True,
|
116 |
+
textbox=gr.MultimodalTextbox(
|
117 |
+
file_count="multiple",
|
118 |
+
file_types=["image"],
|
119 |
+
sources=["upload", "microphone"],
|
120 |
+
),
|
121 |
+
)
|
122 |
+
|
123 |
+
# Lanzar la aplicaci贸n
|
124 |
+
if __name__ == "__main__":
|
125 |
+
demo.launch(debug=True, show_error=True)
|