Version 3.74: Merge latest updates on dev branch (frontier) (#1621)

* Update version to 3.74

* Add support for Yi Model API (#1635)

* 更新以支持零一万物模型

* 删除newbing

* 修改config

---------

Co-authored-by: binary-husky <qingxu.fu@outlook.com>

* Refactor function signatures in bridge files

* fix qwen api change

* rename and ref functions

* rename and move some cookie functions

* 增加haiku模型,新增endpoint配置说明 (#1626)

* haiku added

* 新增haiku,新增endpoint配置说明

* Haiku added

* 将说明同步至最新Endpoint

---------

Co-authored-by: binary-husky <qingxu.fu@outlook.com>

* private_upload目录下进行文件鉴权 (#1596)

* private_upload目录下进行文件鉴权

* minor fastapi adjustment

* Add logging functionality to enable saving
conversation records

* waiting to fix username retrieve

* support 2rd web path

* allow accessing default user dir

---------

Co-authored-by: binary-husky <qingxu.fu@outlook.com>

* remove yaml deps

* fix favicon

* fix abs path auth problem

* forget to write a return

* add `dashscope` to deps

* fix GHSA-v9q9-xj86-953p

* 用户名重叠越权访问patch (#1681)

* add cohere model api access

* cohere + can_multi_thread

* fix block user access(fail)

* fix fastapi bug

* change cohere api endpoint

* explain version

---------

Co-authored-by: Menghuan1918 <menghuan2003@outlook.com>
Co-authored-by: Skyzayre <120616113+Skyzayre@users.noreply.github.com>
Co-authored-by: XIao <46100050+Kilig947@users.noreply.github.com>
This commit is contained in:
binary-husky
2024-04-08 11:49:30 +08:00
committed by GitHub
parent ae51a0e686
commit bdd46c5dd1
28 changed files with 868 additions and 253 deletions

View File

@@ -34,6 +34,9 @@ from .bridge_google_gemini import predict_no_ui_long_connection as genai_noui
from .bridge_zhipu import predict_no_ui_long_connection as zhipu_noui
from .bridge_zhipu import predict as zhipu_ui
from .bridge_cohere import predict as cohere_ui
from .bridge_cohere import predict_no_ui_long_connection as cohere_noui
colors = ['#FF00FF', '#00FFFF', '#FF0000', '#990099', '#009999', '#990044']
class LazyloadTiktoken(object):
@@ -64,6 +67,7 @@ newbing_endpoint = "wss://sydney.bing.com/sydney/ChatHub"
gemini_endpoint = "https://generativelanguage.googleapis.com/v1beta/models"
claude_endpoint = "https://api.anthropic.com/v1/messages"
yimodel_endpoint = "https://api.lingyiwanwu.com/v1/chat/completions"
cohere_endpoint = 'https://api.cohere.ai/v1/chat'
if not AZURE_ENDPOINT.endswith('/'): AZURE_ENDPOINT += '/'
azure_endpoint = AZURE_ENDPOINT + f'openai/deployments/{AZURE_ENGINE}/chat/completions?api-version=2023-05-15'
@@ -82,6 +86,7 @@ if newbing_endpoint in API_URL_REDIRECT: newbing_endpoint = API_URL_REDIRECT[new
if gemini_endpoint in API_URL_REDIRECT: gemini_endpoint = API_URL_REDIRECT[gemini_endpoint]
if claude_endpoint in API_URL_REDIRECT: claude_endpoint = API_URL_REDIRECT[claude_endpoint]
if yimodel_endpoint in API_URL_REDIRECT: yimodel_endpoint = API_URL_REDIRECT[yimodel_endpoint]
if cohere_endpoint in API_URL_REDIRECT: cohere_endpoint = API_URL_REDIRECT[cohere_endpoint]
# 获取tokenizer
tokenizer_gpt35 = LazyloadTiktoken("gpt-3.5-turbo")
@@ -310,6 +315,18 @@ model_info = {
"tokenizer": tokenizer_gpt35,
"token_cnt": get_token_num_gpt35,
},
# cohere
"cohere-command-r-plus": {
"fn_with_ui": cohere_ui,
"fn_without_ui": cohere_noui,
"can_multi_thread": True,
"endpoint": cohere_endpoint,
"max_token": 1024 * 4,
"tokenizer": tokenizer_gpt35,
"token_cnt": get_token_num_gpt35,
},
}
# -=-=-=-=-=-=- 月之暗面 -=-=-=-=-=-=-
from request_llms.bridge_moonshot import predict as moonshot_ui
@@ -359,7 +376,7 @@ for model in AVAIL_LLM_MODELS:
# -=-=-=-=-=-=- 以下部分是新加入的模型,可能附带额外依赖 -=-=-=-=-=-=-
# claude家族
claude_models = ["claude-instant-1.2","claude-2.0","claude-2.1","claude-3-sonnet-20240229","claude-3-opus-20240229"]
claude_models = ["claude-instant-1.2","claude-2.0","claude-2.1","claude-3-haiku-20240307","claude-3-sonnet-20240229","claude-3-opus-20240229"]
if any(item in claude_models for item in AVAIL_LLM_MODELS):
from .bridge_claude import predict_no_ui_long_connection as claude_noui
from .bridge_claude import predict as claude_ui
@@ -393,6 +410,16 @@ if any(item in claude_models for item in AVAIL_LLM_MODELS):
"token_cnt": get_token_num_gpt35,
},
})
model_info.update({
"claude-3-haiku-20240307": {
"fn_with_ui": claude_ui,
"fn_without_ui": claude_noui,
"endpoint": claude_endpoint,
"max_token": 200000,
"tokenizer": tokenizer_gpt35,
"token_cnt": get_token_num_gpt35,
},
})
model_info.update({
"claude-3-sonnet-20240229": {
"fn_with_ui": claude_ui,
@@ -789,7 +816,7 @@ def LLM_CATCH_EXCEPTION(f):
"""
装饰器函数,将错误显示出来
"""
def decorated(inputs, llm_kwargs, history, sys_prompt, observe_window, console_slience):
def decorated(inputs:str, llm_kwargs:dict, history:list, sys_prompt:str, observe_window:list, console_slience:bool):
try:
return f(inputs, llm_kwargs, history, sys_prompt, observe_window, console_slience)
except Exception as e:
@@ -799,9 +826,9 @@ def LLM_CATCH_EXCEPTION(f):
return decorated
def predict_no_ui_long_connection(inputs, llm_kwargs, history, sys_prompt, observe_window=[], console_slience=False):
def predict_no_ui_long_connection(inputs:str, llm_kwargs:dict, history:list, sys_prompt:str, observe_window:list=[], console_slience:bool=False):
"""
发送至LLM等待回复一次性完成不显示中间过程。但内部用stream的方法避免中途网线被掐。
发送至LLM等待回复一次性完成不显示中间过程。但内部(尽可能地)用stream的方法避免中途网线被掐。
inputs
是本次问询的输入
sys_prompt:
@@ -819,7 +846,6 @@ def predict_no_ui_long_connection(inputs, llm_kwargs, history, sys_prompt, obser
model = llm_kwargs['llm_model']
n_model = 1
if '&' not in model:
assert not model.startswith("tgui"), "TGUI不支持函数插件的实现"
# 如果只询问1个大语言模型
method = model_info[model]["fn_without_ui"]
@@ -880,15 +906,22 @@ def predict_no_ui_long_connection(inputs, llm_kwargs, history, sys_prompt, obser
return res
def predict(inputs, llm_kwargs, *args, **kwargs):
def predict(inputs:str, llm_kwargs:dict, *args, **kwargs):
"""
发送至LLM流式获取输出。
用于基础的对话功能。
inputs 是本次问询的输入
top_p, temperature是LLM的内部调优参数
history 是之前的对话列表注意无论是inputs还是history内容太长了都会触发token数量溢出的错误
chatbot 为WebUI中显示的对话列表修改它然后yeild出去可以直接修改对话界面内容
additional_fn代表点击的哪个按钮按钮见functional.py
完整参数列表:
predict(
inputs:str, # 是本次问询的输入
llm_kwargs:dict, # 是LLM的内部调优参数
plugin_kwargs:dict, # 是插件的内部参数
chatbot:ChatBotWithCookies, # 原样传递,负责向用户前端展示对话,兼顾前端状态的功能
history:list=[], # 是之前的对话列表
system_prompt:str='', # 系统静默prompt
stream:bool=True, # 是否流式输出(已弃用)
additional_fn:str=None # 基础功能区按钮的附加功能
):
"""
inputs = apply_gpt_academic_string_mask(inputs, mode="show_llm")