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
328
request_llms/bridge_cohere.py
Normal file
328
request_llms/bridge_cohere.py
Normal file
@@ -0,0 +1,328 @@
|
||||
# 借鉴了 https://github.com/GaiZhenbiao/ChuanhuChatGPT 项目
|
||||
|
||||
"""
|
||||
该文件中主要包含三个函数
|
||||
|
||||
不具备多线程能力的函数：
|
||||
1. predict: 正常对话时使用，具备完备的交互功能，不可多线程
|
||||
|
||||
具备多线程调用能力的函数
|
||||
2. predict_no_ui_long_connection：支持多线程
|
||||
"""
|
||||
|
||||
import json
|
||||
import time
|
||||
import gradio as gr
|
||||
import logging
|
||||
import traceback
|
||||
import requests
|
||||
import importlib
|
||||
import random
|
||||
|
||||
# config_private.py放自己的秘密如API和代理网址
|
||||
# 读取时首先看是否存在私密的config_private配置文件（不受git管控），如果有，则覆盖原config文件
|
||||
from toolbox import get_conf, update_ui, is_any_api_key, select_api_key, what_keys, clip_history
|
||||
from toolbox import trimmed_format_exc, is_the_upload_folder, read_one_api_model_name, log_chat
|
||||
from toolbox import ChatBotWithCookies
|
||||
proxies, TIMEOUT_SECONDS, MAX_RETRY, API_ORG, AZURE_CFG_ARRAY = \
|
||||
get_conf('proxies', 'TIMEOUT_SECONDS', 'MAX_RETRY', 'API_ORG', 'AZURE_CFG_ARRAY')
|
||||
|
||||
timeout_bot_msg = '[Local Message] Request timeout. Network error. Please check proxy settings in config.py.' + \
|
||||
'网络错误，检查代理服务器是否可用，以及代理设置的格式是否正确，格式须是[协议]://[地址]:[端口]，缺一不可。'
|
||||
|
||||
def get_full_error(chunk, stream_response):
|
||||
"""
|
||||
获取完整的从Cohere返回的报错
|
||||
"""
|
||||
while True:
|
||||
try:
|
||||
chunk += next(stream_response)
|
||||
except:
|
||||
break
|
||||
return chunk
|
||||
|
||||
def decode_chunk(chunk):
|
||||
# 提前读取一些信息 （用于判断异常）
|
||||
chunk_decoded = chunk.decode()
|
||||
chunkjson = None
|
||||
has_choices = False
|
||||
choice_valid = False
|
||||
has_content = False
|
||||
has_role = False
|
||||
try:
|
||||
chunkjson = json.loads(chunk_decoded)
|
||||
has_choices = 'choices' in chunkjson
|
||||
if has_choices: choice_valid = (len(chunkjson['choices']) > 0)
|
||||
if has_choices and choice_valid: has_content = ("content" in chunkjson['choices'][0]["delta"])
|
||||
if has_content: has_content = (chunkjson['choices'][0]["delta"]["content"] is not None)
|
||||
if has_choices and choice_valid: has_role = "role" in chunkjson['choices'][0]["delta"]
|
||||
except:
|
||||
pass
|
||||
return chunk_decoded, chunkjson, has_choices, choice_valid, has_content, has_role
|
||||
|
||||
from functools import lru_cache
|
||||
@lru_cache(maxsize=32)
|
||||
def verify_endpoint(endpoint):
|
||||
"""
|
||||
检查endpoint是否可用
|
||||
"""
|
||||
if "你亲手写的api名称" in endpoint:
|
||||
raise ValueError("Endpoint不正确, 请检查AZURE_ENDPOINT的配置! 当前的Endpoint为:" + endpoint)
|
||||
return endpoint
|
||||
|
||||
def predict_no_ui_long_connection(inputs:str, llm_kwargs:dict, history:list=[], sys_prompt:str="", observe_window:list=None, console_slience:bool=False):
|
||||
"""
|
||||
发送，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。
|
||||
inputs：
|
||||
是本次问询的输入
|
||||
sys_prompt:
|
||||
系统静默prompt
|
||||
llm_kwargs：
|
||||
内部调优参数
|
||||
history：
|
||||
是之前的对话列表
|
||||
observe_window = None：
|
||||
用于负责跨越线程传递已经输出的部分，大部分时候仅仅为了fancy的视觉效果，留空即可。observe_window[0]：观测窗。observe_window[1]：看门狗
|
||||
"""
|
||||
watch_dog_patience = 5 # 看门狗的耐心, 设置5秒即可
|
||||
headers, payload = generate_payload(inputs, llm_kwargs, history, system_prompt=sys_prompt, stream=True)
|
||||
retry = 0
|
||||
while True:
|
||||
try:
|
||||
# make a POST request to the API endpoint, stream=False
|
||||
from .bridge_all import model_info
|
||||
endpoint = verify_endpoint(model_info[llm_kwargs['llm_model']]['endpoint'])
|
||||
response = requests.post(endpoint, headers=headers, proxies=proxies,
|
||||
json=payload, stream=True, timeout=TIMEOUT_SECONDS); break
|
||||
except requests.exceptions.ReadTimeout as e:
|
||||
retry += 1
|
||||
traceback.print_exc()
|
||||
if retry > MAX_RETRY: raise TimeoutError
|
||||
if MAX_RETRY!=0: print(f'请求超时，正在重试 ({retry}/{MAX_RETRY}) ……')
|
||||
|
||||
stream_response = response.iter_lines()
|
||||
result = ''
|
||||
json_data = None
|
||||
while True:
|
||||
try: chunk = next(stream_response)
|
||||
except StopIteration:
|
||||
break
|
||||
except requests.exceptions.ConnectionError:
|
||||
chunk = next(stream_response) # 失败了，重试一次？再失败就没办法了。
|
||||
chunk_decoded, chunkjson, has_choices, choice_valid, has_content, has_role = decode_chunk(chunk)
|
||||
if chunkjson['event_type'] == 'stream-start': continue
|
||||
if chunkjson['event_type'] == 'text-generation':
|
||||
result += chunkjson["text"]
|
||||
if not console_slience: print(chunkjson["text"], end='')
|
||||
if observe_window is not None:
|
||||
# 观测窗，把已经获取的数据显示出去
|
||||
if len(observe_window) >= 1:
|
||||
observe_window[0] += chunkjson["text"]
|
||||
# 看门狗，如果超过期限没有喂狗，则终止
|
||||
if len(observe_window) >= 2:
|
||||
if (time.time()-observe_window[1]) > watch_dog_patience:
|
||||
raise RuntimeError("用户取消了程序。")
|
||||
if chunkjson['event_type'] == 'stream-end': break
|
||||
return result
|
||||
|
||||
|
||||
def predict(inputs:str, llm_kwargs:dict, plugin_kwargs:dict, chatbot:ChatBotWithCookies,
|
||||
history:list=[], system_prompt:str='', stream:bool=True, additional_fn:str=None):
|
||||
"""
|
||||
发送至chatGPT，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chatGPT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
|
||||
"""
|
||||
# if is_any_api_key(inputs):
|
||||
# chatbot._cookies['api_key'] = inputs
|
||||
# chatbot.append(("输入已识别为Cohere的api_key", what_keys(inputs)))
|
||||
# yield from update_ui(chatbot=chatbot, history=history, msg="api_key已导入") # 刷新界面
|
||||
# return
|
||||
# elif not is_any_api_key(chatbot._cookies['api_key']):
|
||||
# chatbot.append((inputs, "缺少api_key。\n\n1. 临时解决方案：直接在输入区键入api_key，然后回车提交。\n\n2. 长效解决方案：在config.py中配置。"))
|
||||
# yield from update_ui(chatbot=chatbot, history=history, msg="缺少api_key") # 刷新界面
|
||||
# return
|
||||
|
||||
user_input = inputs
|
||||
if additional_fn is not None:
|
||||
from core_functional import handle_core_functionality
|
||||
inputs, history = handle_core_functionality(additional_fn, inputs, history, chatbot)
|
||||
|
||||
raw_input = inputs
|
||||
# logging.info(f'[raw_input] {raw_input}')
|
||||
chatbot.append((inputs, ""))
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="等待响应") # 刷新界面
|
||||
|
||||
# check mis-behavior
|
||||
if is_the_upload_folder(user_input):
|
||||
chatbot[-1] = (inputs, f"[Local Message] 检测到操作错误！当您上传文档之后，需点击“**函数插件区**”按钮进行处理，请勿点击“提交”按钮或者“基础功能区”按钮。")
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="正常") # 刷新界面
|
||||
time.sleep(2)
|
||||
|
||||
try:
|
||||
headers, payload = generate_payload(inputs, llm_kwargs, history, system_prompt, stream)
|
||||
except RuntimeError as e:
|
||||
chatbot[-1] = (inputs, f"您提供的api-key不满足要求，不包含任何可用于{llm_kwargs['llm_model']}的api-key。您可能选择了错误的模型或请求源。")
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="api-key不满足要求") # 刷新界面
|
||||
return
|
||||
|
||||
# 检查endpoint是否合法
|
||||
try:
|
||||
from .bridge_all import model_info
|
||||
endpoint = verify_endpoint(model_info[llm_kwargs['llm_model']]['endpoint'])
|
||||
except:
|
||||
tb_str = '```\n' + trimmed_format_exc() + '```'
|
||||
chatbot[-1] = (inputs, tb_str)
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="Endpoint不满足要求") # 刷新界面
|
||||
return
|
||||
|
||||
history.append(inputs); history.append("")
|
||||
|
||||
retry = 0
|
||||
while True:
|
||||
try:
|
||||
# make a POST request to the API endpoint, stream=True
|
||||
response = requests.post(endpoint, headers=headers, proxies=proxies,
|
||||
json=payload, stream=True, timeout=TIMEOUT_SECONDS);break
|
||||
except:
|
||||
retry += 1
|
||||
chatbot[-1] = ((chatbot[-1][0], timeout_bot_msg))
|
||||
retry_msg = f"，正在重试 ({retry}/{MAX_RETRY}) ……" if MAX_RETRY > 0 else ""
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="请求超时"+retry_msg) # 刷新界面
|
||||
if retry > MAX_RETRY: raise TimeoutError
|
||||
|
||||
gpt_replying_buffer = ""
|
||||
|
||||
is_head_of_the_stream = True
|
||||
if stream:
|
||||
stream_response = response.iter_lines()
|
||||
while True:
|
||||
try:
|
||||
chunk = next(stream_response)
|
||||
except StopIteration:
|
||||
# 非Cohere官方接口的出现这样的报错，Cohere和API2D不会走这里
|
||||
chunk_decoded = chunk.decode()
|
||||
error_msg = chunk_decoded
|
||||
# 其他情况，直接返回报错
|
||||
chatbot, history = handle_error(inputs, llm_kwargs, chatbot, history, chunk_decoded, error_msg)
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="非Cohere官方接口返回了错误:" + chunk.decode()) # 刷新界面
|
||||
return
|
||||
|
||||
# 提前读取一些信息 （用于判断异常）
|
||||
chunk_decoded, chunkjson, has_choices, choice_valid, has_content, has_role = decode_chunk(chunk)
|
||||
|
||||
if chunkjson:
|
||||
try:
|
||||
if chunkjson['event_type'] == 'stream-start':
|
||||
continue
|
||||
if chunkjson['event_type'] == 'text-generation':
|
||||
gpt_replying_buffer = gpt_replying_buffer + chunkjson["text"]
|
||||
history[-1] = gpt_replying_buffer
|
||||
chatbot[-1] = (history[-2], history[-1])
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="正常") # 刷新界面
|
||||
if chunkjson['event_type'] == 'stream-end':
|
||||
log_chat(llm_model=llm_kwargs["llm_model"], input_str=inputs, output_str=gpt_replying_buffer)
|
||||
history[-1] = gpt_replying_buffer
|
||||
chatbot[-1] = (history[-2], history[-1])
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="正常") # 刷新界面
|
||||
break
|
||||
except Exception as e:
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="Json解析不合常规") # 刷新界面
|
||||
chunk = get_full_error(chunk, stream_response)
|
||||
chunk_decoded = chunk.decode()
|
||||
error_msg = chunk_decoded
|
||||
chatbot, history = handle_error(inputs, llm_kwargs, chatbot, history, chunk_decoded, error_msg)
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg="Json异常" + error_msg) # 刷新界面
|
||||
print(error_msg)
|
||||
return
|
||||
|
||||
def handle_error(inputs, llm_kwargs, chatbot, history, chunk_decoded, error_msg):
|
||||
from .bridge_all import model_info
|
||||
Cohere_website = ' 请登录Cohere查看详情 https://platform.Cohere.com/signup'
|
||||
if "reduce the length" in error_msg:
|
||||
if len(history) >= 2: history[-1] = ""; history[-2] = "" # 清除当前溢出的输入：history[-2] 是本次输入, history[-1] 是本次输出
|
||||
history = clip_history(inputs=inputs, history=history, tokenizer=model_info[llm_kwargs['llm_model']]['tokenizer'],
|
||||
max_token_limit=(model_info[llm_kwargs['llm_model']]['max_token'])) # history至少释放二分之一
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] Reduce the length. 本次输入过长, 或历史数据过长. 历史缓存数据已部分释放, 您可以请再次尝试. (若再次失败则更可能是因为输入过长.)")
|
||||
elif "does not exist" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], f"[Local Message] Model {llm_kwargs['llm_model']} does not exist. 模型不存在, 或者您没有获得体验资格.")
|
||||
elif "Incorrect API key" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] Incorrect API key. Cohere以提供了不正确的API_KEY为由, 拒绝服务. " + Cohere_website)
|
||||
elif "exceeded your current quota" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] You exceeded your current quota. Cohere以账户额度不足为由, 拒绝服务." + Cohere_website)
|
||||
elif "account is not active" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] Your account is not active. Cohere以账户失效为由, 拒绝服务." + Cohere_website)
|
||||
elif "associated with a deactivated account" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] You are associated with a deactivated account. Cohere以账户失效为由, 拒绝服务." + Cohere_website)
|
||||
elif "API key has been deactivated" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] API key has been deactivated. Cohere以账户失效为由, 拒绝服务." + Cohere_website)
|
||||
elif "bad forward key" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] Bad forward key. API2D账户额度不足.")
|
||||
elif "Not enough point" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] Not enough point. API2D账户点数不足.")
|
||||
else:
|
||||
from toolbox import regular_txt_to_markdown
|
||||
tb_str = '```\n' + trimmed_format_exc() + '```'
|
||||
chatbot[-1] = (chatbot[-1][0], f"[Local Message] 异常 \n\n{tb_str} \n\n{regular_txt_to_markdown(chunk_decoded)}")
|
||||
return chatbot, history
|
||||
|
||||
def generate_payload(inputs, llm_kwargs, history, system_prompt, stream):
|
||||
"""
|
||||
整合所有信息，选择LLM模型，生成http请求，为发送请求做准备
|
||||
"""
|
||||
# if not is_any_api_key(llm_kwargs['api_key']):
|
||||
# raise AssertionError("你提供了错误的API_KEY。\n\n1. 临时解决方案：直接在输入区键入api_key，然后回车提交。\n\n2. 长效解决方案：在config.py中配置。")
|
||||
|
||||
api_key = select_api_key(llm_kwargs['api_key'], llm_kwargs['llm_model'])
|
||||
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
"Authorization": f"Bearer {api_key}"
|
||||
}
|
||||
if API_ORG.startswith('org-'): headers.update({"Cohere-Organization": API_ORG})
|
||||
if llm_kwargs['llm_model'].startswith('azure-'):
|
||||
headers.update({"api-key": api_key})
|
||||
if llm_kwargs['llm_model'] in AZURE_CFG_ARRAY.keys():
|
||||
azure_api_key_unshared = AZURE_CFG_ARRAY[llm_kwargs['llm_model']]["AZURE_API_KEY"]
|
||||
headers.update({"api-key": azure_api_key_unshared})
|
||||
|
||||
conversation_cnt = len(history) // 2
|
||||
|
||||
messages = [{"role": "SYSTEM", "message": system_prompt}]
|
||||
if conversation_cnt:
|
||||
for index in range(0, 2*conversation_cnt, 2):
|
||||
what_i_have_asked = {}
|
||||
what_i_have_asked["role"] = "USER"
|
||||
what_i_have_asked["message"] = history[index]
|
||||
what_gpt_answer = {}
|
||||
what_gpt_answer["role"] = "CHATBOT"
|
||||
what_gpt_answer["message"] = history[index+1]
|
||||
if what_i_have_asked["message"] != "":
|
||||
if what_gpt_answer["message"] == "": continue
|
||||
if what_gpt_answer["message"] == timeout_bot_msg: continue
|
||||
messages.append(what_i_have_asked)
|
||||
messages.append(what_gpt_answer)
|
||||
else:
|
||||
messages[-1]['message'] = what_gpt_answer['message']
|
||||
|
||||
model = llm_kwargs['llm_model']
|
||||
if model.startswith('cohere-'): model = model[len('cohere-'):]
|
||||
payload = {
|
||||
"model": model,
|
||||
"message": inputs,
|
||||
"chat_history": messages,
|
||||
"temperature": llm_kwargs['temperature'], # 1.0,
|
||||
"top_p": llm_kwargs['top_p'], # 1.0,
|
||||
"n": 1,
|
||||
"stream": stream,
|
||||
"presence_penalty": 0,
|
||||
"frequency_penalty": 0,
|
||||
}
|
||||
|
||||
return headers,payload
|
||||
|
||||
|
||||
Reference in New Issue
Block a user