Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,92 +1,34 @@
|
|
1 |
import gradio as gr
|
2 |
-
from transformers import
|
3 |
import torch
|
4 |
-
import os
|
5 |
-
import gc
|
6 |
|
7 |
-
|
8 |
-
|
|
|
|
|
9 |
|
10 |
def get_system_prompt():
|
11 |
with open("prompt.txt", "r", encoding="utf-8") as f:
|
12 |
return f.read().strip()
|
13 |
|
14 |
-
def
|
15 |
-
|
|
|
16 |
|
17 |
-
|
18 |
-
|
19 |
|
20 |
-
|
21 |
-
|
22 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
23 |
-
model = AutoModelForCausalLM.from_pretrained(
|
24 |
-
model_name,
|
25 |
-
torch_dtype=torch.float32,
|
26 |
-
trust_remote_code=True,
|
27 |
-
low_cpu_mem_usage=True
|
28 |
-
)
|
29 |
-
|
30 |
-
model = model.to("cpu")
|
31 |
-
|
32 |
-
gc.collect()
|
33 |
-
torch.cuda.empty_cache() if torch.cuda.is_available() else None
|
34 |
-
|
35 |
-
print("✅ Modelo Falcon-7B cargado en CPU")
|
36 |
-
|
37 |
-
def generate_response(user_message):
|
38 |
-
try:
|
39 |
-
load_model_if_needed()
|
40 |
-
|
41 |
-
if not user_message.strip():
|
42 |
-
return "Por favor, escribe una pregunta para que pueda ayudarte."
|
43 |
-
|
44 |
-
system_prompt = get_system_prompt()
|
45 |
-
|
46 |
-
prompt = f"{system_prompt}\n\nUsuario: {user_message}\nBITER:"
|
47 |
-
|
48 |
-
inputs = tokenizer(prompt, return_tensors="pt")
|
49 |
-
|
50 |
-
generation_config = {
|
51 |
-
"max_new_tokens": 400,
|
52 |
-
"temperature": 0.7,
|
53 |
-
"top_p": 0.9,
|
54 |
-
"do_sample": True,
|
55 |
-
"pad_token_id": tokenizer.eos_token_id,
|
56 |
-
"num_return_sequences": 1
|
57 |
-
}
|
58 |
-
|
59 |
-
with torch.no_grad():
|
60 |
-
outputs = model.generate(**inputs, **generation_config)
|
61 |
-
|
62 |
-
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
63 |
-
answer = response.replace(prompt, "").strip()
|
64 |
-
|
65 |
-
gc.collect()
|
66 |
-
torch.cuda.empty_cache() if torch.cuda.is_available() else None
|
67 |
-
|
68 |
-
return answer
|
69 |
-
|
70 |
-
except Exception as e:
|
71 |
-
print(f"Error: {str(e)}")
|
72 |
-
return f"❌ Lo siento, ha ocurrido un error: {str(e)}"
|
73 |
|
|
|
74 |
demo = gr.Interface(
|
75 |
fn=generate_response,
|
76 |
-
inputs=gr.Textbox(
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
outputs=gr.Textbox(label="Respuesta de BITER"),
|
81 |
-
title="BITER - Tu Mentor en Tiempo Real para Decisiones de Negocio",
|
82 |
-
description="BITER es un asistente de IA que responde dudas de emprendedores como si fuera un CEO experimentado.",
|
83 |
-
examples=[
|
84 |
-
["¿Cómo puedo validar mi idea de negocio con poco presupuesto?"],
|
85 |
-
["¿Cuál es la mejor estrategia para conseguir mis primeros clientes?"],
|
86 |
-
["¿Debería invertir en publicidad en redes sociales o en SEO?"]
|
87 |
-
],
|
88 |
-
allow_flagging="never"
|
89 |
)
|
90 |
|
91 |
if __name__ == "__main__":
|
92 |
-
demo.
|
|
|
1 |
import gradio as gr
|
2 |
+
from transformers import T5Tokenizer, T5ForConditionalGeneration
|
3 |
import torch
|
|
|
|
|
4 |
|
5 |
+
# Cargar modelo y tokenizer
|
6 |
+
model_name = "google/flan-t5-base"
|
7 |
+
tokenizer = T5Tokenizer.from_pretrained(model_name)
|
8 |
+
model = T5ForConditionalGeneration.from_pretrained(model_name)
|
9 |
|
10 |
def get_system_prompt():
|
11 |
with open("prompt.txt", "r", encoding="utf-8") as f:
|
12 |
return f.read().strip()
|
13 |
|
14 |
+
def generate_response(user_input):
|
15 |
+
system_prompt = get_system_prompt()
|
16 |
+
full_prompt = f"{system_prompt}\n\nUsuario: {user_input}\nBITER:"
|
17 |
|
18 |
+
inputs = tokenizer(full_prompt, return_tensors="pt")
|
19 |
+
output = model.generate(**inputs, max_new_tokens=200)
|
20 |
|
21 |
+
decoded_output = tokenizer.decode(output[0], skip_special_tokens=True)
|
22 |
+
return decoded_output.strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
+
# Interfaz Gradio para probar el modelo directamente en Hugging Face
|
25 |
demo = gr.Interface(
|
26 |
fn=generate_response,
|
27 |
+
inputs=gr.Textbox(lines=2, placeholder="Escribe tu pregunta..."),
|
28 |
+
outputs=gr.Textbox(),
|
29 |
+
title="BITER - Mentor IA para Emprendedores",
|
30 |
+
description="Respuestas rápidas, estratégicas y en español. Como un CEO que te asesora al instante.",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
)
|
32 |
|
33 |
if __name__ == "__main__":
|
34 |
+
demo.launch()
|