Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,64 +1,81 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
-
|
|
|
|
| 3 |
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
"""
|
| 7 |
-
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
| 8 |
|
|
|
|
| 9 |
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
history: list[tuple[str, str]],
|
| 13 |
-
system_message,
|
| 14 |
-
max_tokens,
|
| 15 |
-
temperature,
|
| 16 |
-
top_p,
|
| 17 |
-
):
|
| 18 |
-
messages = [{"role": "system", "content": system_message}]
|
| 19 |
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
response = ""
|
| 29 |
-
|
| 30 |
-
for message in client.chat_completion(
|
| 31 |
-
messages,
|
| 32 |
-
max_tokens=max_tokens,
|
| 33 |
-
stream=True,
|
| 34 |
-
temperature=temperature,
|
| 35 |
-
top_p=top_p,
|
| 36 |
-
):
|
| 37 |
-
token = message.choices[0].delta.content
|
| 38 |
|
| 39 |
-
|
| 40 |
-
yield response
|
| 41 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
gr.Slider(
|
| 53 |
-
minimum=
|
| 54 |
-
maximum=
|
| 55 |
-
value=
|
| 56 |
-
|
| 57 |
-
label="Top-p (nucleus sampling)",
|
| 58 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
)
|
| 61 |
|
| 62 |
-
|
| 63 |
if __name__ == "__main__":
|
| 64 |
-
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
import torch
|
| 3 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 4 |
|
| 5 |
+
# Nombre correcto del modelo en Hugging Face
|
| 6 |
+
model_name = "deepseek-ai/deepseek-coder-1.3b-instruct"
|
|
|
|
|
|
|
| 7 |
|
| 8 |
+
print("Cargando modelo... Esto puede tomar unos minutos la primera vez.")
|
| 9 |
|
| 10 |
+
# Cargar el tokenizer
|
| 11 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
+
# Cargar el modelo con optimizaciones para recursos limitados
|
| 14 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 15 |
+
model_name,
|
| 16 |
+
torch_dtype=torch.float16, # Usar float16 para ahorrar memoria
|
| 17 |
+
device_map="auto",
|
| 18 |
+
trust_remote_code=True,
|
| 19 |
+
low_cpu_mem_usage=True
|
| 20 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
+
print("隆Modelo cargado exitosamente!")
|
|
|
|
| 23 |
|
| 24 |
+
def generate_response(prompt, max_length=200, temperature=0.7):
|
| 25 |
+
# Formatear el prompt para DeepSeek Coder
|
| 26 |
+
formatted_prompt = f"### Instruction:\n{prompt}\n\n### Response:\n"
|
| 27 |
+
|
| 28 |
+
inputs = tokenizer.encode(formatted_prompt, return_tensors="pt")
|
| 29 |
+
|
| 30 |
+
with torch.no_grad():
|
| 31 |
+
outputs = model.generate(
|
| 32 |
+
inputs,
|
| 33 |
+
max_length=len(inputs[0]) + max_length,
|
| 34 |
+
temperature=temperature,
|
| 35 |
+
do_sample=True,
|
| 36 |
+
pad_token_id=tokenizer.eos_token_id,
|
| 37 |
+
eos_token_id=tokenizer.eos_token_id,
|
| 38 |
+
repetition_penalty=1.1
|
| 39 |
+
)
|
| 40 |
+
|
| 41 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 42 |
+
# Extraer solo la respuesta generada
|
| 43 |
+
response = response.split("### Response:\n")[-1].strip()
|
| 44 |
+
return response
|
| 45 |
|
| 46 |
+
# Interfaz Gradio
|
| 47 |
+
interface = gr.Interface(
|
| 48 |
+
fn=generate_response,
|
| 49 |
+
inputs=[
|
| 50 |
+
gr.Textbox(
|
| 51 |
+
label="Prompt",
|
| 52 |
+
placeholder="Escribe tu pregunta de programaci贸n o c贸digo...",
|
| 53 |
+
lines=3
|
| 54 |
+
),
|
| 55 |
gr.Slider(
|
| 56 |
+
minimum=50,
|
| 57 |
+
maximum=500,
|
| 58 |
+
value=200,
|
| 59 |
+
label="Longitud m谩xima de respuesta"
|
|
|
|
| 60 |
),
|
| 61 |
+
gr.Slider(
|
| 62 |
+
minimum=0.1,
|
| 63 |
+
maximum=2.0,
|
| 64 |
+
value=0.7,
|
| 65 |
+
step=0.1,
|
| 66 |
+
label="Temperatura (creatividad)"
|
| 67 |
+
)
|
| 68 |
],
|
| 69 |
+
outputs=gr.Textbox(label="Respuesta del DeepSeek Coder", lines=10),
|
| 70 |
+
title="馃殌 DeepSeek Coder 1.3B",
|
| 71 |
+
description="Modelo de programaci贸n DeepSeek ejecut谩ndose en Hugging Face Spaces. Perfecto para ayuda con c贸digo, explicaciones y debugging.",
|
| 72 |
+
examples=[
|
| 73 |
+
["Escribe una funci贸n en Python para calcular fibonacci"],
|
| 74 |
+
["驴C贸mo puedo hacer una API REST con FastAPI?"],
|
| 75 |
+
["Explica qu茅 hace este c贸digo: for i in range(10): print(i**2)"],
|
| 76 |
+
["Crea una funci贸n JavaScript para validar emails"]
|
| 77 |
+
]
|
| 78 |
)
|
| 79 |
|
|
|
|
| 80 |
if __name__ == "__main__":
|
| 81 |
+
interface.launch()
|