yang/CodeGeeX4
1
0
Fork 0
mirror of https://github.com/JasonYANG170/CodeGeeX4.git synced 2024-11-23 12:16:33 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

fix pep8 error

Browse Source
This commit is contained in:
XingYu-Zhong 2024-07-09 11:37:30 +08:00
parent 48fff1449b
commit 25870dc0a1
7 changed files with 289 additions and 262 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
repodemo/llm/api/codegeex4.py
View File

@ -2,40 +2,38 @@ import requests
import json
URL = "" # the url you deploy codegeex service
def codegeex4(prompt, temperature=0.8, top_p=0.8):
url = URL
headers = {
'Content-Type': 'application/json'
}
headers = {"Content-Type": "application/json"}
data = {
'inputs': prompt,
'parameters': {
'best_of':1,
'do_sample': True,
'max_new_tokens': 4012,
'temperature': temperature,
'top_p': top_p,
'stop': ["<|endoftext|>", "<|user|>", "<|observation|>", "<|assistant|>"],
}
"inputs": prompt,
"parameters": {
"best_of": 1,
"do_sample": True,
"max_new_tokens": 4012,
"temperature": temperature,
"top_p": top_p,
"stop": ["<|endoftext|>", "<|user|>", "<|observation|>", "<|assistant|>"],
},
}
response = requests.post(url, json=data, headers=headers, verify=False, stream=True)
if response.status_code == 200:
for line in response.iter_lines():
if line:
decoded_line = line.decode('utf-8').replace('data:', '').strip()
decoded_line = line.decode("utf-8").replace("data:", "").strip()
if decoded_line:
try:
content = json.loads(decoded_line)
token_text = content.get('token', {}).get('text', '')
if '<|endoftext|>' in token_text:
token_text = content.get("token", {}).get("text", "")
if "<|endoftext|>" in token_text:
break
yield token_text
except json.JSONDecodeError:
continue
else:
print('请求失败:', response.status_code)
print("请求失败:", response.status_code)

22
repodemo/llm/local/codegeex4.py
View File

@ -3,7 +3,8 @@ from transformers import AutoModel, AutoTokenizer
from typing import Iterator
import torch
class CodegeexChatModel():
class CodegeexChatModel:
device: str = Field(description="device to load the model")
tokenizer = Field(description="model's tokenizer")
model = Field(description="Codegeex model")
@ -12,11 +13,14 @@ class CodegeexChatModel():
def __init__(self, model_name_or_path):
super().__init__()
self.device = "cuda" if torch.cuda.is_available() else "cpu"
self.tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=True)
self.model = AutoModel.from_pretrained(
model_name_or_path,
trust_remote_code=True
).to(self.device).eval()
self.tokenizer = AutoTokenizer.from_pretrained(
model_name_or_path, trust_remote_code=True
)
self.model = (
AutoModel.from_pretrained(model_name_or_path, trust_remote_code=True)
.to(self.device)
.eval()
)
print("Model has been initialized.")
def chat(self, prompt, temperature=0.2, top_p=0.95):
@ -26,7 +30,7 @@ class CodegeexChatModel():
query=prompt,
max_length=120000,
temperature=temperature,
top_p=top_p
top_p=top_p,
)
return response
except Exception as e:
@ -40,8 +44,8 @@ class CodegeexChatModel():
query=prompt,
max_length=120000,
temperature=temperature,
top_p=top_p
top_p=top_p,
):
yield response
except Exception as e:
yield f'error: {e}'
yield f"error: {e}"

13
repodemo/prompts/base_prompt.py
View File

@ -19,18 +19,21 @@ web_search_prompy = """
- 除了代码和特定的名称和引用外，您的答案必须使用与问题相同的语言来撰写。
""".lstrip()
def get_cur_base_user_prompt(message_history, index_prompt=None, judge_context=""):
user_prompt_tmp = """<|user|>\n{user_input}"""
assistant_prompt_tmp = """<|assistant|>\n{assistant_input}"""
history_prompt = ""
for i, message in enumerate(message_history):
if message['role'] == 'user':
if message["role"] == "user":
if i == 0 and index_prompt is not None:
history_prompt += "<|user|>\n"+index_prompt+message['content']
history_prompt += "<|user|>\n" + index_prompt + message["content"]
else:
history_prompt += user_prompt_tmp.format(user_input=message['content'])
elif message['role'] == 'assistant':
history_prompt += assistant_prompt_tmp.format(assistant_input=message['content'])
history_prompt += user_prompt_tmp.format(user_input=message["content"])
elif message["role"] == "assistant":
history_prompt += assistant_prompt_tmp.format(
assistant_input=message["content"]
)
# print("修改" not in judge_context)
# print(judge_context)

74
repodemo/run.py
View File

@ -1,7 +1,11 @@
import chainlit as cl
from chainlit.input_widget import Slider
from llm.api.codegeex4 import codegeex4
from prompts.base_prompt import judge_task_prompt,get_cur_base_user_prompt,web_judge_task_prompt
from prompts.base_prompt import (
judge_task_prompt,
get_cur_base_user_prompt,
web_judge_task_prompt,
)
from utils.tools import unzip_file, get_project_files_with_content
from utils.bingsearch import bing_search_prompt
@ -16,37 +20,29 @@ async def chat_profile():
cl.Starter(
label="请你用python写一个快速排序。",
message="请你用python写一个快速排序。",
),
cl.Starter(
label="请你介绍一下自己。",
message="请你介绍一下自己。",
),
cl.Starter(
label="用 Python 编写一个脚本来自动发送每日电子邮件报告，并指导我如何进行设置。",
message="用 Python 编写一个脚本来自动发送每日电子邮件报告，并指导我如何进行设置。",
),
cl.Starter(
label="我是一个python初学者，请你告诉我怎么才能学好python。",
message="我是一个python初学者，请你告诉我怎么才能学好python。",
)
]
),
],
),
cl.ChatProfile(
name="联网问答",
markdown_description="联网能力dome：支持联网回答用户问题。",
markdown_description="联网能力demo：支持联网回答用户问题。",
),
cl.ChatProfile(
name="上传本地项目",
markdown_description="项目级能力dome：支持上传本地zip压缩包项目，可以进行项目问答和对项目进行修改。",
)
markdown_description="项目级能力demo：支持上传本地zip压缩包项目，可以进行项目问答和对项目进行修改。",
),
]
@ -74,21 +70,20 @@ async def start():
).send()
temperature = settings["temperature"]
top_p = settings["top_p"]
cl.user_session.set('temperature',temperature)
cl.user_session.set('top_p',top_p)
cl.user_session.set(
"message_history",
[]
)
cl.user_session.set("temperature", temperature)
cl.user_session.set("top_p", top_p)
cl.user_session.set("message_history", [])
chat_profile = cl.user_session.get("chat_profile")
extract_dir = 'repodata'
extract_dir = "repodata"
if chat_profile == "chat聊天":
pass
elif chat_profile == "上传本地项目":
files = None
while files == None:
files = await cl.AskFileMessage(
content="请上传项目zip压缩文件!", accept={"application/zip": [".zip"]},max_size_mb=50
content="请上传项目zip压缩文件!",
accept={"application/zip": [".zip"]},
max_size_mb=50,
).send()
text_file = files[0]
@ -101,7 +96,6 @@ async def start():
).send()
@cl.on_message
async def main(message: cl.Message):
chat_profile = cl.user_session.get("chat_profile")
@ -111,12 +105,16 @@ async def main(message: cl.Message):
prompt_content = get_cur_base_user_prompt(message_history=message_history)
elif chat_profile == "联网问答":
judge_tmp = codegeex4(web_judge_task_prompt.format(user_input=message.content),temperature=0.2,top_p = 0.95)
judge_context = '\n'.join(judge_tmp)
judge_tmp = codegeex4(
web_judge_task_prompt.format(user_input=message.content),
temperature=0.2,
top_p=0.95,
)
judge_context = "\n".join(judge_tmp)
print(judge_context)
message_history.pop()
if '是' in judge_context:
if "是" in judge_context:
prompt_tmp = bing_search_prompt(message.content)
message_history.append({"role": "user", "content": prompt_tmp})
else:
@ -124,8 +122,12 @@ async def main(message: cl.Message):
prompt_content = get_cur_base_user_prompt(message_history=message_history)
elif chat_profile == "上传本地项目":
judge_tmp = codegeex4(judge_task_prompt.format(user_input=message.content),temperature=0.2,top_p = 0.95)
judge_context = ''
judge_tmp = codegeex4(
judge_task_prompt.format(user_input=message.content),
temperature=0.2,
top_p=0.95,
)
judge_context = ""
for part in judge_tmp:
judge_context += part
@ -133,16 +135,22 @@ async def main(message: cl.Message):
index_prompt = ""
index_tmp = """###PATH:{path}\n{code}\n"""
for index in project_index:
index_prompt+=index_tmp.format(path=index['path'],code=index['content'])
index_prompt += index_tmp.format(path=index["path"], code=index["content"])
print(judge_context)
prompt_content = get_cur_base_user_prompt(message_history=message_history,index_prompt=index_prompt,judge_context=judge_context) if '正常' not in judge_context else get_cur_base_user_prompt(message_history=message_history)
prompt_content = (
get_cur_base_user_prompt(
message_history=message_history,
index_prompt=index_prompt,
judge_context=judge_context,
)
if "正常" not in judge_context
else get_cur_base_user_prompt(message_history=message_history)
)
msg = cl.Message(content="")
await msg.send()
temperature = cl.user_session.get("temperature")
top_p = cl.user_session.get('top_p')
top_p = cl.user_session.get("top_p")
if len(prompt_content) / 4 < 120000:
stream = codegeex4(prompt_content, temperature=temperature, top_p=top_p)

74
repodemo/run_local.py
View File

@ -1,13 +1,19 @@
import chainlit as cl
from chainlit.input_widget import Slider
from llm.api.codegeex4 import codegeex4
from prompts.base_prompt import judge_task_prompt,get_cur_base_user_prompt,web_judge_task_prompt
from prompts.base_prompt import (
judge_task_prompt,
get_cur_base_user_prompt,
web_judge_task_prompt,
)
from utils.tools import unzip_file, get_project_files_with_content
from utils.bingsearch import bing_search_prompt
from llm.local.codegeex4 import CodegeexChatModel
local_model_path = '<your_local_model_path>'
local_model_path = "<your_local_model_path>"
llm = CodegeexChatModel(local_model_path)
class StreamProcessor:
def __init__(self):
self.previous_str = ""
@ -17,6 +23,7 @@ class StreamProcessor:
self.previous_str = new_str
return new_part
@cl.set_chat_profiles
async def chat_profile():
return [
@ -27,37 +34,29 @@ async def chat_profile():
cl.Starter(
label="请你用python写一个快速排序。",
message="请你用python写一个快速排序。",
),
cl.Starter(
label="请你介绍一下自己。",
message="请你介绍一下自己。",
),
cl.Starter(
label="用 Python 编写一个脚本来自动发送每日电子邮件报告，并指导我如何进行设置。",
message="用 Python 编写一个脚本来自动发送每日电子邮件报告，并指导我如何进行设置。",
),
cl.Starter(
label="我是一个python初学者，请你告诉我怎么才能学好python。",
message="我是一个python初学者，请你告诉我怎么才能学好python。",
)
]
),
],
),
cl.ChatProfile(
name="联网问答",
markdown_description="联网能力dome：支持联网回答用户问题。",
markdown_description="联网能力demo：支持联网回答用户问题。",
),
cl.ChatProfile(
name="上传本地项目",
markdown_description="项目级能力dome：支持上传本地zip压缩包项目，可以进行项目问答和对项目进行修改。",
)
markdown_description="项目级能力demo：支持上传本地zip压缩包项目，可以进行项目问答和对项目进行修改。",
),
]
@ -85,21 +84,20 @@ async def start():
).send()
temperature = settings["temperature"]
top_p = settings["top_p"]
cl.user_session.set('temperature',temperature)
cl.user_session.set('top_p',top_p)
cl.user_session.set(
"message_history",
[]
)
cl.user_session.set("temperature", temperature)
cl.user_session.set("top_p", top_p)
cl.user_session.set("message_history", [])
chat_profile = cl.user_session.get("chat_profile")
extract_dir = 'repodata'
extract_dir = "repodata"
if chat_profile == "chat聊天":
pass
elif chat_profile == "上传本地项目":
files = None
while files == None:
files = await cl.AskFileMessage(
content="请上传项目zip压缩文件!", accept={"application/zip": [".zip"]},max_size_mb=50
content="请上传项目zip压缩文件!",
accept={"application/zip": [".zip"]},
max_size_mb=50,
).send()
text_file = files[0]
@ -112,7 +110,6 @@ async def start():
).send()
@cl.on_message
async def main(message: cl.Message):
chat_profile = cl.user_session.get("chat_profile")
@ -122,11 +119,13 @@ async def main(message: cl.Message):
prompt_content = get_cur_base_user_prompt(message_history=message_history)
elif chat_profile == "联网问答":
judge_context = llm.chat(web_judge_task_prompt.format(user_input=message.content),temperature=0.2)
judge_context = llm.chat(
web_judge_task_prompt.format(user_input=message.content), temperature=0.2
)
print(judge_context)
message_history.pop()
if '是' in judge_context:
if "是" in judge_context:
prompt_tmp = bing_search_prompt(message.content)
message_history.append({"role": "user", "content": prompt_tmp})
else:
@ -134,23 +133,30 @@ async def main(message: cl.Message):
prompt_content = get_cur_base_user_prompt(message_history=message_history)
elif chat_profile == "上传本地项目":
judge_context = llm.chat(judge_task_prompt.format(user_input=message.content),temperature=0.2)
judge_context = llm.chat(
judge_task_prompt.format(user_input=message.content), temperature=0.2
)
project_index = cl.user_session.get("project_index")
index_prompt = ""
index_tmp = """###PATH:{path}\n{code}\n"""
for index in project_index:
index_prompt+=index_tmp.format(path=index['path'],code=index['content'])
index_prompt += index_tmp.format(path=index["path"], code=index["content"])
print(judge_context)
prompt_content = get_cur_base_user_prompt(message_history=message_history,index_prompt=index_prompt,judge_context=judge_context) if '正常' not in judge_context else get_cur_base_user_prompt(message_history=message_history)
prompt_content = (
get_cur_base_user_prompt(
message_history=message_history,
index_prompt=index_prompt,
judge_context=judge_context,
)
if "正常" not in judge_context
else get_cur_base_user_prompt(message_history=message_history)
)
msg = cl.Message(content="")
await msg.send()
temperature = cl.user_session.get("temperature")
top_p = cl.user_session.get('top_p')
top_p = cl.user_session.get("top_p")
if len(prompt_content) / 4 < 120000:
stream = llm.stream_chat(prompt_content, temperature=temperature, top_p=top_p)
@ -159,7 +165,7 @@ async def main(message: cl.Message):
if isinstance(part, str):
text = stream_processor.get_new_part(part)
elif isinstance(part, dict):
text = stream_processor.get_new_part(part['name']+part['content'])
text = stream_processor.get_new_part(part["name"] + part["content"])
if token := (text or " "):
await msg.stream_token(token)
else:

20
repodemo/utils/bingsearch.py
View File

@ -2,7 +2,9 @@ import requests
from bs4 import BeautifulSoup as BS4
import requests
BING_API_KEY = '<your_bing_api_key>'
BING_API_KEY = "<your_bing_api_key>"
def search_with_bing(query: str, search_timeout=30, top_k=6) -> list[dict]:
"""
Search with bing and return the contexts.
@ -13,9 +15,9 @@ def search_with_bing(query: str, search_timeout=30, top_k=6) -> list[dict]:
headers={"Ocp-Apim-Subscription-Key": BING_API_KEY},
params={
"q": query,
"responseFilter": ['webpages'],
"freshness": 'month',
"mkt": 'zh-CN'
"responseFilter": ["webpages"],
"freshness": "month",
"mkt": "zh-CN",
},
timeout=search_timeout,
)
@ -30,18 +32,22 @@ def search_with_bing(query: str, search_timeout=30, top_k=6) -> list[dict]:
print(f"搜索失败，错误原因: {e}")
return []
def fetch_url(url):
response = requests.get(url)
# use beautifulsoup4 to parse html
soup = BS4(response.text, 'html.parser')
soup = BS4(response.text, "html.parser")
plain_text = soup.get_text()
return plain_text
def bing_search_prompt(input):
contents = search_with_bing(input, search_timeout=5, top_k=6)
citations = "\n\n".join(
[f"[[citation:{i + 1}]]\n```markdown\n{item['snippet']}\n```" for i, item in enumerate(contents)]
[
f"[[citation:{i + 1}]]\n```markdown\n{item['snippet']}\n```"
for i, item in enumerate(contents)
]
)
prompt = f"[引用]\n{citations}\n问：{input}\n"
return prompt

18
repodemo/utils/tools.py
View File

@ -2,6 +2,7 @@ import zipfile
import os
import json
def unzip_file(zip_path, extract_dir):
"""
解压zip文件到指定目录，并在指定目录下创建一个新的目录存放解压后的文件
@ -23,7 +24,7 @@ def unzip_file(zip_path, extract_dir):
if not os.path.exists(new_extract_dir):
os.makedirs(new_extract_dir)
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
with zipfile.ZipFile(zip_path, "r") as zip_ref:
zip_ref.extractall(new_extract_dir)
return new_extract_dir
@ -48,14 +49,15 @@ def get_project_files_with_content(project_dir):
relative_path = os.path.relpath(file_path, project_dir)
if "__MACOSX" in relative_path:
continue
with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
with open(file_path, "r", encoding="utf-8", errors="ignore") as f:
content = f.read()
files_list.append({'path': relative_path, 'content': content})
files_list.append({"path": relative_path, "content": content})
else:
continue
return files_list
def filter_data(obj):
LANGUAGE_TAG = {
"c++": "// C++",
@ -89,7 +91,6 @@ def filter_data(obj):
"tex": f"% TeX",
"typescript": "// TypeScript",
"vue": "<!-- Vue -->",
"assembly": "; Assembly",
"dart": "// Dart",
"perl": "# Perl",
@ -124,7 +125,9 @@ def filter_data(obj):
"dockerfile": "# Dockerfile",
}
programming_languages_to_file_extensions = json.load(open('utils/programming-languages-to-file-extensions.json'))
programming_languages_to_file_extensions = json.load(
open("utils/programming-languages-to-file-extensions.json")
)
need2del = []
for key in programming_languages_to_file_extensions.keys():
if key.lower() not in LANGUAGE_TAG:
@ -140,8 +143,8 @@ def filter_data(obj):
ext_to_programming_languages[item] = key
want_languages.append(item)
ext = '.'+obj.split('.')[-1]
with open('utils/keep.txt', 'r') as f:
ext = "." + obj.split(".")[-1]
with open("utils/keep.txt", "r") as f:
keep_files = f.readlines()
keep_files = [l.strip() for l in keep_files]
# print(ext)
@ -151,4 +154,3 @@ def filter_data(obj):
return False
else:
return True