version 3.75 (#1702)
* 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 * # fix com_zhipuglm.py illegal temperature problem (#1687) * Update com_zhipuglm.py # fix 用户在使用 zhipuai 界面时遇到了关于温度参数的非法参数错误 * allow store lm model dropdown * add a btn to reverse previous reset * remove extra fns * Add support for glm-4v model (#1700) * 修改chatglm3量化加载方式 (#1688) Co-authored-by: zym9804 <ren990603@gmail.com> * save chat stage 1 * consider null cookie situation * 在点击复制按钮时激活语音 * miss some parts * move all to js * done first stage * add edge tts * bug fix * bug fix * remove console log * bug fix * bug fix * bug fix * audio switch * update tts readme * remove tempfile when done * disable auto audio follow * avoid play queue update after shut up * feat: minimizing common.js * improve tts functionality * deterine whether the cached model is in choices * Add support for Ollama (#1740) * print err when doc2x not successful * add icon * adjust url for doc2x key version * prepare merge --------- 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> Co-authored-by: Yuki <903728862@qq.com> Co-authored-by: zyren123 <91042213+zyren123@users.noreply.github.com> Co-authored-by: zym9804 <ren990603@gmail.com>
This commit is contained in:
binary-husky
@@ -67,7 +67,8 @@ 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'
|
||||
cohere_endpoint = "https://api.cohere.ai/v1/chat"
|
||||
ollama_endpoint = "http://localhost:11434/api/chat"
|
||||
|
||||
if not AZURE_ENDPOINT.endswith('/'): AZURE_ENDPOINT += '/'
|
||||
azure_endpoint = AZURE_ENDPOINT + f'openai/deployments/{AZURE_ENGINE}/chat/completions?api-version=2023-05-15'
|
||||
@@ -87,6 +88,7 @@ if gemini_endpoint in API_URL_REDIRECT: gemini_endpoint = API_URL_REDIRECT[gemin
|
||||
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]
|
||||
if ollama_endpoint in API_URL_REDIRECT: ollama_endpoint = API_URL_REDIRECT[ollama_endpoint]
|
||||
|
||||
# 获取tokenizer
|
||||
tokenizer_gpt35 = LazyloadTiktoken("gpt-3.5-turbo")
|
||||
@@ -266,6 +268,14 @@ model_info = {
|
||||
"tokenizer": tokenizer_gpt35,
|
||||
"token_cnt": get_token_num_gpt35,
|
||||
},
|
||||
"glm-4v": {
|
||||
"fn_with_ui": zhipu_ui,
|
||||
"fn_without_ui": zhipu_noui,
|
||||
"endpoint": None,
|
||||
"max_token": 1000,
|
||||
"tokenizer": tokenizer_gpt35,
|
||||
"token_cnt": get_token_num_gpt35,
|
||||
},
|
||||
"glm-3-turbo": {
|
||||
"fn_with_ui": zhipu_ui,
|
||||
"fn_without_ui": zhipu_noui,
|
||||
@@ -827,7 +837,32 @@ for model in [m for m in AVAIL_LLM_MODELS if m.startswith("vllm-")]:
|
||||
"token_cnt": get_token_num_gpt35,
|
||||
},
|
||||
})
|
||||
|
||||
# -=-=-=-=-=-=- ollama 对齐支持 -=-=-=-=-=-=-
|
||||
for model in [m for m in AVAIL_LLM_MODELS if m.startswith("ollama-")]:
|
||||
from .bridge_ollama import predict_no_ui_long_connection as ollama_noui
|
||||
from .bridge_ollama import predict as ollama_ui
|
||||
break
|
||||
for model in [m for m in AVAIL_LLM_MODELS if m.startswith("ollama-")]:
|
||||
# 为了更灵活地接入ollama多模型管理界面，设计了此接口，例子：AVAIL_LLM_MODELS = ["ollama-phi3(max_token=6666)"]
|
||||
# 其中
|
||||
# "ollama-" 是前缀（必要）
|
||||
# "phi3" 是模型名（必要）
|
||||
# "(max_token=6666)" 是配置（非必要）
|
||||
try:
|
||||
_, max_token_tmp = read_one_api_model_name(model)
|
||||
except:
|
||||
print(f"ollama模型 {model} 的 max_token 配置不是整数，请检查配置文件。")
|
||||
continue
|
||||
model_info.update({
|
||||
model: {
|
||||
"fn_with_ui": ollama_ui,
|
||||
"fn_without_ui": ollama_noui,
|
||||
"endpoint": ollama_endpoint,
|
||||
"max_token": max_token_tmp,
|
||||
"tokenizer": tokenizer_gpt35,
|
||||
"token_cnt": get_token_num_gpt35,
|
||||
},
|
||||
})
|
||||
|
||||
# -=-=-=-=-=-=- azure模型对齐支持 -=-=-=-=-=-=-
|
||||
AZURE_CFG_ARRAY = get_conf("AZURE_CFG_ARRAY") # <-- 用于定义和切换多个azure模型 -->
|
||||
|
||||
@@ -6,7 +6,6 @@ from toolbox import get_conf, ProxyNetworkActivate
|
||||
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
|
||||
|
||||
|
||||
|
||||
# ------------------------------------------------------------------------------------------------------------------------
|
||||
# 🔌💻 Local Model
|
||||
# ------------------------------------------------------------------------------------------------------------------------
|
||||
@@ -23,20 +22,45 @@ class GetGLM3Handle(LocalLLMHandle):
|
||||
import os, glob
|
||||
import os
|
||||
import platform
|
||||
LOCAL_MODEL_QUANT, device = get_conf('LOCAL_MODEL_QUANT', 'LOCAL_MODEL_DEVICE')
|
||||
|
||||
if LOCAL_MODEL_QUANT == "INT4": # INT4
|
||||
_model_name_ = "THUDM/chatglm3-6b-int4"
|
||||
elif LOCAL_MODEL_QUANT == "INT8": # INT8
|
||||
_model_name_ = "THUDM/chatglm3-6b-int8"
|
||||
else:
|
||||
_model_name_ = "THUDM/chatglm3-6b" # FP16
|
||||
with ProxyNetworkActivate('Download_LLM'):
|
||||
chatglm_tokenizer = AutoTokenizer.from_pretrained(_model_name_, trust_remote_code=True)
|
||||
if device=='cpu':
|
||||
chatglm_model = AutoModel.from_pretrained(_model_name_, trust_remote_code=True, device='cpu').float()
|
||||
LOCAL_MODEL_QUANT, device = get_conf("LOCAL_MODEL_QUANT", "LOCAL_MODEL_DEVICE")
|
||||
_model_name_ = "THUDM/chatglm3-6b"
|
||||
# if LOCAL_MODEL_QUANT == "INT4": # INT4
|
||||
# _model_name_ = "THUDM/chatglm3-6b-int4"
|
||||
# elif LOCAL_MODEL_QUANT == "INT8": # INT8
|
||||
# _model_name_ = "THUDM/chatglm3-6b-int8"
|
||||
# else:
|
||||
# _model_name_ = "THUDM/chatglm3-6b" # FP16
|
||||
with ProxyNetworkActivate("Download_LLM"):
|
||||
chatglm_tokenizer = AutoTokenizer.from_pretrained(
|
||||
_model_name_, trust_remote_code=True
|
||||
)
|
||||
if device == "cpu":
|
||||
chatglm_model = AutoModel.from_pretrained(
|
||||
_model_name_,
|
||||
trust_remote_code=True,
|
||||
device="cpu",
|
||||
).float()
|
||||
elif LOCAL_MODEL_QUANT == "INT4": # INT4
|
||||
chatglm_model = AutoModel.from_pretrained(
|
||||
pretrained_model_name_or_path=_model_name_,
|
||||
trust_remote_code=True,
|
||||
device="cuda",
|
||||
load_in_4bit=True,
|
||||
)
|
||||
elif LOCAL_MODEL_QUANT == "INT8": # INT8
|
||||
chatglm_model = AutoModel.from_pretrained(
|
||||
pretrained_model_name_or_path=_model_name_,
|
||||
trust_remote_code=True,
|
||||
device="cuda",
|
||||
load_in_8bit=True,
|
||||
)
|
||||
else:
|
||||
chatglm_model = AutoModel.from_pretrained(_model_name_, trust_remote_code=True, device='cuda')
|
||||
chatglm_model = AutoModel.from_pretrained(
|
||||
pretrained_model_name_or_path=_model_name_,
|
||||
trust_remote_code=True,
|
||||
device="cuda",
|
||||
)
|
||||
chatglm_model = chatglm_model.eval()
|
||||
|
||||
self._model = chatglm_model
|
||||
@@ -46,32 +70,36 @@ class GetGLM3Handle(LocalLLMHandle):
|
||||
def llm_stream_generator(self, **kwargs):
|
||||
# 🏃♂️🏃♂️🏃♂️ 子进程执行
|
||||
def adaptor(kwargs):
|
||||
query = kwargs['query']
|
||||
max_length = kwargs['max_length']
|
||||
top_p = kwargs['top_p']
|
||||
temperature = kwargs['temperature']
|
||||
history = kwargs['history']
|
||||
query = kwargs["query"]
|
||||
max_length = kwargs["max_length"]
|
||||
top_p = kwargs["top_p"]
|
||||
temperature = kwargs["temperature"]
|
||||
history = kwargs["history"]
|
||||
return query, max_length, top_p, temperature, history
|
||||
|
||||
query, max_length, top_p, temperature, history = adaptor(kwargs)
|
||||
|
||||
for response, history in self._model.stream_chat(self._tokenizer,
|
||||
query,
|
||||
history,
|
||||
max_length=max_length,
|
||||
top_p=top_p,
|
||||
temperature=temperature,
|
||||
):
|
||||
for response, history in self._model.stream_chat(
|
||||
self._tokenizer,
|
||||
query,
|
||||
history,
|
||||
max_length=max_length,
|
||||
top_p=top_p,
|
||||
temperature=temperature,
|
||||
):
|
||||
yield response
|
||||
|
||||
def try_to_import_special_deps(self, **kwargs):
|
||||
# import something that will raise error if the user does not install requirement_*.txt
|
||||
# 🏃♂️🏃♂️🏃♂️ 主进程执行
|
||||
import importlib
|
||||
|
||||
# importlib.import_module('modelscope')
|
||||
|
||||
|
||||
# ------------------------------------------------------------------------------------------------------------------------
|
||||
# 🔌💻 GPT-Academic Interface
|
||||
# ------------------------------------------------------------------------------------------------------------------------
|
||||
predict_no_ui_long_connection, predict = get_local_llm_predict_fns(GetGLM3Handle, model_name, history_format='chatglm3')
|
||||
predict_no_ui_long_connection, predict = get_local_llm_predict_fns(
|
||||
GetGLM3Handle, model_name, history_format="chatglm3"
|
||||
)
|
||||
|
||||
272
request_llms/bridge_ollama.py
Normal file
272
request_llms/bridge_ollama.py
Normal file
@@ -0,0 +1,272 @@
|
||||
# 借鉴自同目录下的bridge_chatgpt.py
|
||||
|
||||
"""
|
||||
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, trimmed_format_exc, is_the_upload_folder, read_one_api_model_name
|
||||
proxies, TIMEOUT_SECONDS, MAX_RETRY = get_conf(
|
||||
"proxies", "TIMEOUT_SECONDS", "MAX_RETRY"
|
||||
)
|
||||
|
||||
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Openai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
|
||||
is_last_chunk = False
|
||||
try:
|
||||
chunkjson = json.loads(chunk_decoded)
|
||||
is_last_chunk = chunkjson.get("done", False)
|
||||
except:
|
||||
pass
|
||||
return chunk_decoded, chunkjson, is_last_chunk
|
||||
|
||||
def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=None, console_slience=False):
|
||||
"""
|
||||
发送至chatGPT，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：
|
||||
chatGPT的内部调优参数
|
||||
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
|
||||
if inputs == "": inputs = "空空如也的输入栏"
|
||||
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 = 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 = ''
|
||||
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, is_last_chunk = decode_chunk(chunk)
|
||||
if chunk:
|
||||
try:
|
||||
if is_last_chunk:
|
||||
# 判定为数据流的结束，gpt_replying_buffer也写完了
|
||||
logging.info(f'[response] {result}')
|
||||
break
|
||||
result += chunkjson['message']["content"]
|
||||
if not console_slience: print(chunkjson['message']["content"], 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['message']["content"]
|
||||
# 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。")
|
||||
except Exception as e:
|
||||
chunk = get_full_error(chunk, stream_response)
|
||||
chunk_decoded = chunk.decode()
|
||||
error_msg = chunk_decoded
|
||||
print(error_msg)
|
||||
raise RuntimeError("Json解析不合常规")
|
||||
return result
|
||||
|
||||
|
||||
def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=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 inputs == "": inputs = "空空如也的输入栏"
|
||||
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)
|
||||
|
||||
headers, payload = generate_payload(inputs, llm_kwargs, history, system_prompt, stream)
|
||||
|
||||
from .bridge_all import model_info
|
||||
endpoint = model_info[llm_kwargs['llm_model']]['endpoint']
|
||||
|
||||
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 = ""
|
||||
|
||||
if stream:
|
||||
stream_response = response.iter_lines()
|
||||
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。
|
||||
|
||||
# 提前读取一些信息 （用于判断异常）
|
||||
chunk_decoded, chunkjson, is_last_chunk = decode_chunk(chunk)
|
||||
|
||||
if chunk:
|
||||
try:
|
||||
if is_last_chunk:
|
||||
# 判定为数据流的结束，gpt_replying_buffer也写完了
|
||||
logging.info(f'[response] {gpt_replying_buffer}')
|
||||
break
|
||||
# 处理数据流的主体
|
||||
try:
|
||||
status_text = f"finish_reason: {chunkjson['error'].get('message', 'null')}"
|
||||
except:
|
||||
status_text = "finish_reason: null"
|
||||
gpt_replying_buffer = gpt_replying_buffer + chunkjson['message']["content"]
|
||||
# 如果这里抛出异常，一般是文本过长，详情见get_full_error的输出
|
||||
history[-1] = gpt_replying_buffer
|
||||
chatbot[-1] = (history[-2], history[-1])
|
||||
yield from update_ui(chatbot=chatbot, history=history, msg=status_text) # 刷新界面
|
||||
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
|
||||
if "bad_request" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] 已经超过了模型的最大上下文或是模型格式错误,请尝试削减单次输入的文本量。")
|
||||
elif "authentication_error" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] Incorrect API key. 请确保API key有效。")
|
||||
elif "not_found" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], f"[Local Message] {llm_kwargs['llm_model']} 无效，请确保使用小写的模型名称。")
|
||||
elif "rate_limit" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] 遇到了控制请求速率限制，请一分钟后重试。")
|
||||
elif "system_busy" in error_msg:
|
||||
chatbot[-1] = (chatbot[-1][0], "[Local Message] 系统繁忙，请一分钟后重试。")
|
||||
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
|
||||
"""
|
||||
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
}
|
||||
|
||||
conversation_cnt = len(history) // 2
|
||||
|
||||
messages = [{"role": "system", "content": 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["content"] = history[index]
|
||||
what_gpt_answer = {}
|
||||
what_gpt_answer["role"] = "assistant"
|
||||
what_gpt_answer["content"] = history[index+1]
|
||||
if what_i_have_asked["content"] != "":
|
||||
if what_gpt_answer["content"] == "": continue
|
||||
if what_gpt_answer["content"] == timeout_bot_msg: continue
|
||||
messages.append(what_i_have_asked)
|
||||
messages.append(what_gpt_answer)
|
||||
else:
|
||||
messages[-1]['content'] = what_gpt_answer['content']
|
||||
|
||||
what_i_ask_now = {}
|
||||
what_i_ask_now["role"] = "user"
|
||||
what_i_ask_now["content"] = inputs
|
||||
messages.append(what_i_ask_now)
|
||||
model = llm_kwargs['llm_model']
|
||||
if llm_kwargs['llm_model'].startswith('ollama-'):
|
||||
model = llm_kwargs['llm_model'][len('ollama-'):]
|
||||
model, _ = read_one_api_model_name(model)
|
||||
options = {"temperature": llm_kwargs['temperature']}
|
||||
payload = {
|
||||
"model": model,
|
||||
"messages": messages,
|
||||
"options": options,
|
||||
}
|
||||
try:
|
||||
print(f" {llm_kwargs['llm_model']} : {conversation_cnt} : {inputs[:100]} ..........")
|
||||
except:
|
||||
print('输入中可能存在乱码。')
|
||||
return headers,payload
|
||||
@@ -75,6 +75,10 @@ def predict(inputs:str, llm_kwargs:dict, plugin_kwargs:dict, chatbot:ChatBotWith
|
||||
llm_kwargs["llm_model"] = zhipuai_default_model
|
||||
|
||||
if llm_kwargs["llm_model"] in ["glm-4v"]:
|
||||
if (len(inputs) + sum(len(temp) for temp in history) + 1047) > 2000:
|
||||
chatbot.append((inputs, "上下文长度超过glm-4v上限2000tokens，注意图片大约占用1,047个tokens"))
|
||||
yield from update_ui(chatbot=chatbot, history=history)
|
||||
return
|
||||
have_recent_file, image_paths = have_any_recent_upload_image_files(chatbot)
|
||||
if not have_recent_file:
|
||||
chatbot.append((inputs, "没有检测到任何近期上传的图像文件，请上传jpg格式的图片，此外，请注意拓展名需要小写"))
|
||||
|
||||
@@ -36,8 +36,14 @@ class ZhipuChatInit:
|
||||
what_i_have_asked = {"role": "user", "content": []}
|
||||
what_i_have_asked['content'].append({"type": 'text', "text": user_input})
|
||||
if encode_img:
|
||||
if len(encode_img) > 1:
|
||||
logging.warning("glm-4v只支持一张图片,将只取第一张图片进行处理")
|
||||
print("glm-4v只支持一张图片,将只取第一张图片进行处理")
|
||||
img_d = {"type": "image_url",
|
||||
"image_url": {'url': encode_img}}
|
||||
"image_url": {
|
||||
"url": encode_img[0]['data']
|
||||
}
|
||||
}
|
||||
what_i_have_asked['content'].append(img_d)
|
||||
return what_i_have_asked
|
||||
|
||||
|
||||
Reference in New Issue
Block a user