File size: 5,382 Bytes
3d04cf1
9a33833
98bd34e
e8aafa4
3d04cf1
735d156
3d04cf1
 
 
 
 
7b1af97
 
 
 
 
 
 
3d04cf1
 
 
 
 
 
 
96a7392
3d04cf1
 
 
 
96a7392
 
 
 
 
 
 
 
 
 
 
 
 
 
3d04cf1
 
 
 
 
96a7392
3d04cf1
 
96a7392
 
 
 
3d04cf1
96a7392
3d04cf1
 
96a7392
3d04cf1
 
 
6220229
 
 
 
 
 
 
3d04cf1
 
 
 
 
6220229
7882172
3d04cf1
 
 
 
 
6220229
7882172
3d04cf1
 
 
 
 
 
 
6220229
7882172
9d8a1d3
3d04cf1
6220229
7882172
3d04cf1
 
6220229
7882172
3d04cf1
 
9d8a1d3
7882172
3d04cf1
 
 
 
 
9d8a1d3
9243ca2
3719586
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
95569ad
3719586
95569ad
 
3719586
3d04cf1
 
3e028aa
3d04cf1
 
3e028aa
3d04cf1
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
# app.py

import gradio as gr
import os
from models import load_embedding_model, load_yi_coder_model
from pinecone_utils import connect_to_pinecone, vector_search  # Ahora debería funcionar correctamente
from ui import build_interface
from config import SIMILARITY_THRESHOLD_DEFAULT, SYSTEM_PROMPT, MAX_LENGTH_DEFAULT
from decorators import gpu_decorator
import torch

########################

from utils import process_tags_chat

########################


# Cargar modelos
embedding_model = load_embedding_model()
tokenizer, yi_coder_model, yi_coder_device = load_yi_coder_model()

# Conectar a Pinecone
index = connect_to_pinecone()

# Función para generar código utilizando Yi-Coder
@gpu_decorator(duration=100)
def generate_code(system_prompt, user_prompt, max_length):
    device = yi_coder_device
    model = yi_coder_model
    tokenizer_ = tokenizer  # Ya lo tenemos cargado

    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": user_prompt}
    ]
    
    # Aplicar la plantilla de chat y preparar el texto
    text = tokenizer_.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    model_inputs = tokenizer_([text], return_tensors="pt").to(device)

    with torch.no_grad():
        generated_ids = model.generate(
            model_inputs.input_ids,
            max_new_tokens=max_length,
            eos_token_id=tokenizer_.eos_token_id  
        )

    # Extraer solo la parte generada
    generated_ids = [
        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
    ]

    response = tokenizer_.batch_decode(generated_ids, skip_special_tokens=True)[0]
    return response


# Función para combinar búsqueda vectorial y Yi-Coder
@gpu_decorator(duration=100)
def combined_function(user_prompt, similarity_threshold, selected_option, system_prompt, max_length):
    def get_partial_message(response):
        """Obtiene el contenido después de 'Respuesta:' si está presente en la respuesta."""
        if "Respuesta:" in response:
            return response.split("Respuesta:")[1].strip()  # Tomar solo el texto después de 'Respuesta:'
        else:
            return response  # Devolver la respuesta completa si no contiene 'Respuesta:'

    if selected_option == "Solo Búsqueda Vectorial":
        # Realizar búsqueda vectorial
        search_results = vector_search(user_prompt, embedding_model, index)
        if search_results:
            content = search_results[0]['content']
            partial_message = get_partial_message(content)
            return partial_message, None
        else:
            return "No se encontraron resultados en Pinecone.", None
    elif selected_option == "Solo Yi-Coder":
        # Generar respuesta usando Yi-Coder
        yi_coder_response = generate_code(system_prompt, user_prompt, max_length)
        partial_message = get_partial_message(yi_coder_response)
        return partial_message, None
    elif selected_option == "Ambos (basado en umbral de similitud)":
        # Realizar búsqueda vectorial
        search_results = vector_search(user_prompt, embedding_model, index)
        if search_results:
            top_result = search_results[0]
            if top_result['score'] >= similarity_threshold:
                content = top_result['content']
                partial_message = get_partial_message(content)
                return partial_message, None
            else:
                yi_coder_response = generate_code(system_prompt, user_prompt, max_length)
                partial_message = get_partial_message(yi_coder_response)
                return partial_message, None
        else:
            yi_coder_response = generate_code(system_prompt, user_prompt, max_length)
            partial_message = get_partial_message(yi_coder_response)
            return partial_message, None
    else:
        return "Opción no válida.", None


# Funciones para el procesamiento de entradas y actualización de imágenes
def process_input(message, history, selected_option, similarity_threshold, system_prompt, max_length):
    response, image = combined_function(message, similarity_threshold, selected_option, system_prompt, max_length)
    history.append((message, response))
    return history, history, image

def update_image(image_url):
    """
    Retorna los datos binarios de la imagen para ser mostrados en Gradio.

    Args:
        image_url (str): Ruta de la imagen.

    Returns:
        bytes o None: Datos binarios de la imagen si existe, de lo contrario None.
    """
    if image_url and os.path.exists(image_url):
        try:
            with open(image_url, "rb") as img_file:
                image_data = img_file.read()
            return image_data
        except Exception as e:
            print(f"Error al leer la imagen: {e}")
            return None
    else:
        print("No se encontró una imagen válida.")
        return None


def send_preset_question(question, history, selected_option, similarity_threshold, system_prompt, max_length):
    return process_input(question, history, selected_option, similarity_threshold, system_prompt, max_length)

# Construir y lanzar la interfaz
demo = build_interface(process_input, send_preset_question, update_image)

if __name__ == "__main__":
    demo.launch()