Vladislav Krasnov
commited on
Commit
·
c1fbe89
1
Parent(s):
2aa5988
Update space 6
Browse files
app.py
CHANGED
|
@@ -1,84 +1,80 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 3 |
import torch
|
|
|
|
| 4 |
|
| 5 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
model_name = "microsoft/phi-2"
|
|
|
|
| 7 |
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
| 8 |
if tokenizer.pad_token is None:
|
| 9 |
tokenizer.pad_token = tokenizer.eos_token
|
| 10 |
|
|
|
|
| 11 |
model = AutoModelForCausalLM.from_pretrained(
|
| 12 |
model_name,
|
| 13 |
torch_dtype=torch.float32,
|
| 14 |
device_map="cpu",
|
| 15 |
trust_remote_code=True
|
| 16 |
)
|
|
|
|
| 17 |
|
| 18 |
-
def
|
| 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 |
-
chatbot
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
def user_message(message, history):
|
| 60 |
-
return "", history + [[message, None]]
|
| 61 |
-
|
| 62 |
-
def bot_message(history):
|
| 63 |
-
message = history[-1][0]
|
| 64 |
-
# Вызываем функцию respond
|
| 65 |
-
new_history, _, _ = respond(message, history[:-1])
|
| 66 |
-
history[-1][1] = new_history[-1][1]
|
| 67 |
-
return history
|
| 68 |
-
|
| 69 |
-
# Привязка событий
|
| 70 |
-
msg.submit(user_message, [msg, chatbot], [msg, chatbot], queue=False).then(
|
| 71 |
-
bot_message, chatbot, chatbot
|
| 72 |
-
)
|
| 73 |
-
|
| 74 |
-
clear.click(lambda: None, None, chatbot, queue=False)
|
| 75 |
-
|
| 76 |
-
# Выводим информацию о запуске
|
| 77 |
-
print("=" * 20)
|
| 78 |
-
print(f"Space name: {YOUR_SPACE_NAME}")
|
| 79 |
-
print(f"Username: {YOUR_USERNAME}")
|
| 80 |
-
print(f"Local URL: http://0.0.0.0:7860")
|
| 81 |
-
print(f"Public API endpoint: https://{YOUR_USERNAME}-{YOUR_SPACE_NAME}.hf.space/run/predict")
|
| 82 |
-
print("=" * 20)
|
| 83 |
|
| 84 |
-
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 3 |
import torch
|
| 4 |
+
import os
|
| 5 |
|
| 6 |
+
# ЗАМЕНИТЕ ЭТИ ЗНАЧЕНИЯ НА СВОИ!
|
| 7 |
+
SPACE_NAME = "livecoder" # Имя вашего Space
|
| 8 |
+
USERNAME = "sarekuwa" # Ваш username на Hugging Face
|
| 9 |
+
|
| 10 |
+
# Выводим endpoint сразу при запуске
|
| 11 |
+
print("=" * 50)
|
| 12 |
+
print(f"Space name: {SPACE_NAME}")
|
| 13 |
+
print(f"Username: {USERNAME}")
|
| 14 |
+
print(f"Local URL: http://0.0.0.0:7860")
|
| 15 |
+
print(f"Public API endpoint: https://{USERNAME}-{SPACE_NAME}.hf.space/run/predict")
|
| 16 |
+
print("=" * 50)
|
| 17 |
+
|
| 18 |
+
# Загрузка модели
|
| 19 |
model_name = "microsoft/phi-2"
|
| 20 |
+
print("Loading tokenizer...")
|
| 21 |
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
| 22 |
if tokenizer.pad_token is None:
|
| 23 |
tokenizer.pad_token = tokenizer.eos_token
|
| 24 |
|
| 25 |
+
print("Loading model...")
|
| 26 |
model = AutoModelForCausalLM.from_pretrained(
|
| 27 |
model_name,
|
| 28 |
torch_dtype=torch.float32,
|
| 29 |
device_map="cpu",
|
| 30 |
trust_remote_code=True
|
| 31 |
)
|
| 32 |
+
print("Model loaded successfully!")
|
| 33 |
|
| 34 |
+
def predict(message, history):
|
| 35 |
+
"""Обработчик сообщений"""
|
| 36 |
+
try:
|
| 37 |
+
# Формируем промпт
|
| 38 |
+
prompt = ""
|
| 39 |
+
if history:
|
| 40 |
+
for human_msg, ai_msg in history:
|
| 41 |
+
prompt += f"### Human: {human_msg}\n### Assistant: {ai_msg}\n"
|
| 42 |
+
prompt += f"### Human: {message}\n### Assistant:"
|
| 43 |
+
|
| 44 |
+
# Токенизация и генерация
|
| 45 |
+
inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1024)
|
| 46 |
+
|
| 47 |
+
with torch.no_grad():
|
| 48 |
+
outputs = model.generate(
|
| 49 |
+
inputs.input_ids,
|
| 50 |
+
max_new_tokens=256,
|
| 51 |
+
temperature=0.7,
|
| 52 |
+
do_sample=True,
|
| 53 |
+
top_p=0.9,
|
| 54 |
+
pad_token_id=tokenizer.pad_token_id,
|
| 55 |
+
eos_token_id=tokenizer.eos_token_id
|
| 56 |
+
)
|
| 57 |
+
|
| 58 |
+
response = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
|
| 59 |
+
response = response.split("### Human:")[0].strip()
|
| 60 |
+
return response
|
| 61 |
+
|
| 62 |
+
except Exception as e:
|
| 63 |
+
return f"Ошибка: {str(e)}"
|
| 64 |
|
| 65 |
+
# Исправляем предупреждение: добавляем type='messages'
|
| 66 |
+
demo = gr.ChatInterface(
|
| 67 |
+
fn=predict,
|
| 68 |
+
title="LiveCoder LLM API",
|
| 69 |
+
description="Модель Phi-2 для помощи в написании кода",
|
| 70 |
+
chatbot=gr.Chatbot(height=400, type='messages'), # ИСПРАВЛЕНО ЗДЕСЬ
|
| 71 |
+
textbox=gr.Textbox(placeholder="Введите вопрос по программированию...", lines=3),
|
| 72 |
+
examples=[
|
| 73 |
+
["Напиши hello world на C++"],
|
| 74 |
+
["Объясни принцип ООП"],
|
| 75 |
+
["Как работает указатель в C++?"]
|
| 76 |
+
]
|
| 77 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
+
if __name__ == "__main__":
|
| 80 |
+
demo.launch(server_name="0.0.0.0", server_port=7860, share=False)
|