chunzhi/chatgpt_academic
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

merge jittor branch

Browse Source
This commit is contained in:
binary-husky 2023-05-06 23:39:57 +08:00
parent 62d14cfa3f
commit 4b9078a9dc
7 changed files with 573 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
docs/Dockerfile+JittorLLM Normal file
View File

@ -0,0 +1,59 @@
# How to build | 如何构建: docker build -t gpt-academic-jittor --network=host -f Dockerfile+ChatGLM .
# How to run | (1) 我想直接一键运行（选择0号GPU）: docker run --rm -it --net=host --gpus \"device=0\" gpt-academic-jittor bash
# How to run | (2) 我想运行之前进容器做一些调整（选择1号GPU）: docker run --rm -it --net=host --gpus \"device=1\" gpt-academic-jittor bash
# 从NVIDIA源，从而支持显卡运损（检查宿主的nvidia-smi中的cuda版本必须>=11.3）
FROM nvidia/cuda:11.3.1-runtime-ubuntu20.04
ARG useProxyNetwork=''
RUN apt-get update
RUN apt-get install -y curl proxychains curl g++
RUN apt-get install -y git python python3 python-dev python3-dev --fix-missing
# 配置代理网络（构建Docker镜像时使用）
# # comment out below if you do not need proxy network | 如果不需要翻墙 - 从此行向下删除
RUN $useProxyNetwork curl cip.cc
RUN sed -i '$ d' /etc/proxychains.conf
RUN sed -i '$ d' /etc/proxychains.conf
# 在这里填写主机的代理协议（用于从github拉取代码）
RUN echo "socks5 127.0.0.1 10880" >> /etc/proxychains.conf
ARG useProxyNetwork=proxychains
# # comment out above if you do not need proxy network | 如果不需要翻墙 - 从此行向上删除
# use python3 as the system default python
RUN curl -sS https://bootstrap.pypa.io/get-pip.py | python3.8
# 下载pytorch
RUN $useProxyNetwork python3 -m pip install torch --extra-index-url https://download.pytorch.org/whl/cu113
# 下载分支
WORKDIR /gpt
RUN $useProxyNetwork git clone https://github.com/binary-husky/chatgpt_academic.git -b jittor
WORKDIR /gpt/chatgpt_academic
RUN $useProxyNetwork python3 -m pip install -r requirements.txt
RUN $useProxyNetwork python3 -m pip install -r request_llm/requirements_chatglm.txt
RUN $useProxyNetwork python3 -m pip install -r request_llm/requirements_newbing.txt
RUN $useProxyNetwork python3 -m pip install -r request_llm/requirements_jittorllms.txt -i https://pypi.jittor.org/simple -I
# 下载JittorLLMs
RUN $useProxyNetwork git clone https://github.com/binary-husky/JittorLLMs.git --depth 1 request_llm/jittorllms
# 禁用缓存，确保更新代码
ADD "https://www.random.org/cgi-bin/randbyte?nbytes=10&format=h" skipcache
RUN $useProxyNetwork git pull
# 预热Tiktoken模块
RUN python3 -c 'from check_proxy import warm_up_modules; warm_up_modules()'
# 为chatgpt-academic配置代理和API-KEY （非必要 可选步骤）
# 可同时填写多个API-KEY，支持openai的key和api2d的key共存，用英文逗号分割，例如API_KEY = "sk-openaikey1,fkxxxx-api2dkey2,........"
# LLM_MODEL 是选择初始的模型
# LOCAL_MODEL_DEVICE 是选择chatglm等本地模型运行的设备，可选 cpu 和 cuda
# [说明: 以下内容与`config.py`一一对应，请查阅config.py来完成一下配置的填写]
RUN echo ' \n\
API_KEY = "sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx,fkxxxxxx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx" \n\
USE_PROXY = True \n\
LLM_MODEL = "chatglm" \n\
LOCAL_MODEL_DEVICE = "cuda" \n\
proxies = { "http": "socks5h://localhost:10880", "https": "socks5h://localhost:10880", } ' >> config_private.py
# 启动
CMD ["python3", "-u", "main.py"]

45
request_llm/bridge_all.py
View File

@ -133,6 +133,51 @@ model_info = {
}
AVAIL_LLM_MODELS, = get_conf("AVAIL_LLM_MODELS")
if "jittorllms_rwkv" in AVAIL_LLM_MODELS:
from .bridge_jittorllms_rwkv import predict_no_ui_long_connection as rwkv_noui
from .bridge_jittorllms_rwkv import predict as rwkv_ui
model_info.update({
"jittorllms_rwkv": {
"fn_with_ui": rwkv_ui,
"fn_without_ui": rwkv_noui,
"endpoint": None,
"max_token": 1024,
"tokenizer": tokenizer_gpt35,
"token_cnt": get_token_num_gpt35,
},
})
if "jittorllms_llama" in AVAIL_LLM_MODELS:
from .bridge_jittorllms_llama import predict_no_ui_long_connection as llama_noui
from .bridge_jittorllms_llama import predict as llama_ui
model_info.update({
"jittorllms_llama": {
"fn_with_ui": llama_ui,
"fn_without_ui": llama_noui,
"endpoint": None,
"max_token": 1024,
"tokenizer": tokenizer_gpt35,
"token_cnt": get_token_num_gpt35,
},
})
if "jittorllms_pangualpha" in AVAIL_LLM_MODELS:
from .bridge_jittorllms_pangualpha import predict_no_ui_long_connection as pangualpha_noui
from .bridge_jittorllms_pangualpha import predict as pangualpha_ui
model_info.update({
"jittorllms_pangualpha": {
"fn_with_ui": pangualpha_ui,
"fn_without_ui": pangualpha_noui,
"endpoint": None,
"max_token": 1024,
"tokenizer": tokenizer_gpt35,
"token_cnt": get_token_num_gpt35,
},
})
def LLM_CATCH_EXCEPTION(f):
"""
装饰器函数，将错误显示出来

178
request_llm/bridge_jittorllms_llama.py Normal file
View File

@ -0,0 +1,178 @@
from transformers import AutoModel, AutoTokenizer
import time
import threading
import importlib
from toolbox import update_ui, get_conf
from multiprocessing import Process, Pipe
load_message = "jittorllms尚未加载，加载需要一段时间。注意，请避免混用多种jittor模型，否则可能导致显存溢出而造成卡顿，取决于`config.py`的配置，jittorllms消耗大量的内存（CPU）或显存（GPU），也许会导致低配计算机卡死 ……"
#################################################################################
class GetGLMHandle(Process):
def __init__(self):
super().__init__(daemon=True)
self.parent, self.child = Pipe()
self.jittorllms_model = None
self.info = ""
self.local_history = []
self.success = True
self.check_dependency()
self.start()
self.threadLock = threading.Lock()
def check_dependency(self):
try:
import pandas
self.info = "依赖检测通过"
self.success = True
except:
from toolbox import trimmed_format_exc
self.info = r"缺少jittorllms的依赖，如果要使用jittorllms，除了基础的pip依赖以外，您还需要运行`pip install -r request_llm/requirements_jittorllms.txt -i https://pypi.jittor.org/simple -I`"+\
r"和`git clone https://gitlink.org.cn/jittor/JittorLLMs.git --depth 1 request_llm/jittorllms`两个指令来安装jittorllms的依赖（在项目根目录运行这两个指令）。" +\
r"警告：安装jittorllms依赖后将完全破坏现有的pytorch环境，建议使用docker环境！" + trimmed_format_exc()
self.success = False
def ready(self):
return self.jittorllms_model is not None
def run(self):
# 子进程执行
# 第一次运行，加载参数
def validate_path():
import os, sys
dir_name = os.path.dirname(__file__)
env = os.environ.get("PATH", "")
os.environ["PATH"] = env.replace('/cuda/bin', '/x/bin')
root_dir_assume = os.path.abspath(os.path.dirname(__file__) + '/..')
os.chdir(root_dir_assume + '/request_llm/jittorllms')
sys.path.append(root_dir_assume + '/request_llm/jittorllms')
validate_path() # validate path so you can run from base directory
def load_model():
import types
try:
if self.jittorllms_model is None:
device, = get_conf('LOCAL_MODEL_DEVICE')
from .jittorllms.models import get_model
# availabel_models = ["chatglm", "pangualpha", "llama", "chatrwkv"]
args_dict = {'model': 'llama'}
print('self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))')
self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))
print('done get model')
except:
self.child.send('[Local Message] Call jittorllms fail 不能正常加载jittorllms的参数。')
raise RuntimeError("不能正常加载jittorllms的参数！")
print('load_model')
load_model()
# 进入任务等待状态
print('进入任务等待状态')
while True:
# 进入任务等待状态
kwargs = self.child.recv()
query = kwargs['query']
history = kwargs['history']
# 是否重置
if len(self.local_history) > 0 and len(history)==0:
print('触发重置')
self.jittorllms_model.reset()
self.local_history.append(query)
print('收到消息，开始请求')
try:
for response in self.jittorllms_model.stream_chat(query, history):
print(response)
self.child.send(response)
except:
from toolbox import trimmed_format_exc
print(trimmed_format_exc())
self.child.send('[Local Message] Call jittorllms fail.')
# 请求处理结束，开始下一个循环
self.child.send('[Finish]')
def stream_chat(self, **kwargs):
# 主进程执行
self.threadLock.acquire()
self.parent.send(kwargs)
while True:
res = self.parent.recv()
if res != '[Finish]':
yield res
else:
break
self.threadLock.release()
global llama_glm_handle
llama_glm_handle = None
#################################################################################
def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=[], console_slience=False):
"""
多线程方法
函数的说明请见 request_llm/bridge_all.py
"""
global llama_glm_handle
if llama_glm_handle is None:
llama_glm_handle = GetGLMHandle()
if len(observe_window) >= 1: observe_window[0] = load_message + "\n\n" + llama_glm_handle.info
if not llama_glm_handle.success:
error = llama_glm_handle.info
llama_glm_handle = None
raise RuntimeError(error)
# jittorllms 没有 sys_prompt 接口，因此把prompt加入 history
history_feedin = []
for i in range(len(history)//2):
history_feedin.append([history[2*i], history[2*i+1]] )
watch_dog_patience = 5 # 看门狗 (watchdog) 的耐心, 设置5秒即可
response = ""
for response in llama_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=sys_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
print(response)
if len(observe_window) >= 1: observe_window[0] = response
if len(observe_window) >= 2:
if (time.time()-observe_window[1]) > watch_dog_patience:
raise RuntimeError("程序终止。")
return response
def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=None):
"""
单线程方法
函数的说明请见 request_llm/bridge_all.py
"""
chatbot.append((inputs, ""))
global llama_glm_handle
if llama_glm_handle is None:
llama_glm_handle = GetGLMHandle()
chatbot[-1] = (inputs, load_message + "\n\n" + llama_glm_handle.info)
yield from update_ui(chatbot=chatbot, history=[])
if not llama_glm_handle.success:
llama_glm_handle = None
return
if additional_fn is not None:
import core_functional
importlib.reload(core_functional) # 热更新prompt
core_functional = core_functional.get_core_functions()
if "PreProcess" in core_functional[additional_fn]: inputs = core_functional[additional_fn]["PreProcess"](inputs) # 获取预处理函数（如果有的话）
inputs = core_functional[additional_fn]["Prefix"] + inputs + core_functional[additional_fn]["Suffix"]
# 处理历史信息
history_feedin = []
for i in range(len(history)//2):
history_feedin.append([history[2*i], history[2*i+1]] )
# 开始接收jittorllms的回复
response = "[Local Message]: 等待jittorllms响应中 ..."
for response in llama_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=system_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
chatbot[-1] = (inputs, response)
yield from update_ui(chatbot=chatbot, history=history)
# 总结输出
if response == "[Local Message]: 等待jittorllms响应中 ...":
response = "[Local Message]: jittorllms响应异常 ..."
history.extend([inputs, response])
yield from update_ui(chatbot=chatbot, history=history)

178
request_llm/bridge_jittorllms_pangualpha.py Normal file
View File

@ -0,0 +1,178 @@
from transformers import AutoModel, AutoTokenizer
import time
import threading
import importlib
from toolbox import update_ui, get_conf
from multiprocessing import Process, Pipe
load_message = "jittorllms尚未加载，加载需要一段时间。注意，请避免混用多种jittor模型，否则可能导致显存溢出而造成卡顿，取决于`config.py`的配置，jittorllms消耗大量的内存（CPU）或显存（GPU），也许会导致低配计算机卡死 ……"
#################################################################################
class GetGLMHandle(Process):
def __init__(self):
super().__init__(daemon=True)
self.parent, self.child = Pipe()
self.jittorllms_model = None
self.info = ""
self.local_history = []
self.success = True
self.check_dependency()
self.start()
self.threadLock = threading.Lock()
def check_dependency(self):
try:
import pandas
self.info = "依赖检测通过"
self.success = True
except:
from toolbox import trimmed_format_exc
self.info = r"缺少jittorllms的依赖，如果要使用jittorllms，除了基础的pip依赖以外，您还需要运行`pip install -r request_llm/requirements_jittorllms.txt -i https://pypi.jittor.org/simple -I`"+\
r"和`git clone https://gitlink.org.cn/jittor/JittorLLMs.git --depth 1 request_llm/jittorllms`两个指令来安装jittorllms的依赖（在项目根目录运行这两个指令）。" +\
r"警告：安装jittorllms依赖后将完全破坏现有的pytorch环境，建议使用docker环境！" + trimmed_format_exc()
self.success = False
def ready(self):
return self.jittorllms_model is not None
def run(self):
# 子进程执行
# 第一次运行，加载参数
def validate_path():
import os, sys
dir_name = os.path.dirname(__file__)
env = os.environ.get("PATH", "")
os.environ["PATH"] = env.replace('/cuda/bin', '/x/bin')
root_dir_assume = os.path.abspath(os.path.dirname(__file__) + '/..')
os.chdir(root_dir_assume + '/request_llm/jittorllms')
sys.path.append(root_dir_assume + '/request_llm/jittorllms')
validate_path() # validate path so you can run from base directory
def load_model():
import types
try:
if self.jittorllms_model is None:
device, = get_conf('LOCAL_MODEL_DEVICE')
from .jittorllms.models import get_model
# availabel_models = ["chatglm", "pangualpha", "llama", "chatrwkv"]
args_dict = {'model': 'pangualpha'}
print('self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))')
self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))
print('done get model')
except:
self.child.send('[Local Message] Call jittorllms fail 不能正常加载jittorllms的参数。')
raise RuntimeError("不能正常加载jittorllms的参数！")
print('load_model')
load_model()
# 进入任务等待状态
print('进入任务等待状态')
while True:
# 进入任务等待状态
kwargs = self.child.recv()
query = kwargs['query']
history = kwargs['history']
# 是否重置
if len(self.local_history) > 0 and len(history)==0:
print('触发重置')
self.jittorllms_model.reset()
self.local_history.append(query)
print('收到消息，开始请求')
try:
for response in self.jittorllms_model.stream_chat(query, history):
print(response)
self.child.send(response)
except:
from toolbox import trimmed_format_exc
print(trimmed_format_exc())
self.child.send('[Local Message] Call jittorllms fail.')
# 请求处理结束，开始下一个循环
self.child.send('[Finish]')
def stream_chat(self, **kwargs):
# 主进程执行
self.threadLock.acquire()
self.parent.send(kwargs)
while True:
res = self.parent.recv()
if res != '[Finish]':
yield res
else:
break
self.threadLock.release()
global pangu_glm_handle
pangu_glm_handle = None
#################################################################################
def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=[], console_slience=False):
"""
多线程方法
函数的说明请见 request_llm/bridge_all.py
"""
global pangu_glm_handle
if pangu_glm_handle is None:
pangu_glm_handle = GetGLMHandle()
if len(observe_window) >= 1: observe_window[0] = load_message + "\n\n" + pangu_glm_handle.info
if not pangu_glm_handle.success:
error = pangu_glm_handle.info
pangu_glm_handle = None
raise RuntimeError(error)
# jittorllms 没有 sys_prompt 接口，因此把prompt加入 history
history_feedin = []
for i in range(len(history)//2):
history_feedin.append([history[2*i], history[2*i+1]] )
watch_dog_patience = 5 # 看门狗 (watchdog) 的耐心, 设置5秒即可
response = ""
for response in pangu_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=sys_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
print(response)
if len(observe_window) >= 1: observe_window[0] = response
if len(observe_window) >= 2:
if (time.time()-observe_window[1]) > watch_dog_patience:
raise RuntimeError("程序终止。")
return response
def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=None):
"""
单线程方法
函数的说明请见 request_llm/bridge_all.py
"""
chatbot.append((inputs, ""))
global pangu_glm_handle
if pangu_glm_handle is None:
pangu_glm_handle = GetGLMHandle()
chatbot[-1] = (inputs, load_message + "\n\n" + pangu_glm_handle.info)
yield from update_ui(chatbot=chatbot, history=[])
if not pangu_glm_handle.success:
pangu_glm_handle = None
return
if additional_fn is not None:
import core_functional
importlib.reload(core_functional) # 热更新prompt
core_functional = core_functional.get_core_functions()
if "PreProcess" in core_functional[additional_fn]: inputs = core_functional[additional_fn]["PreProcess"](inputs) # 获取预处理函数（如果有的话）
inputs = core_functional[additional_fn]["Prefix"] + inputs + core_functional[additional_fn]["Suffix"]
# 处理历史信息
history_feedin = []
for i in range(len(history)//2):
history_feedin.append([history[2*i], history[2*i+1]] )
# 开始接收jittorllms的回复
response = "[Local Message]: 等待jittorllms响应中 ..."
for response in pangu_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=system_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
chatbot[-1] = (inputs, response)
yield from update_ui(chatbot=chatbot, history=history)
# 总结输出
if response == "[Local Message]: 等待jittorllms响应中 ...":
response = "[Local Message]: jittorllms响应异常 ..."
history.extend([inputs, response])
yield from update_ui(chatbot=chatbot, history=history)

81
request_llm/bridge_jittorllms.py → request_llm/bridge_jittorllms_rwkv.py
View File

@ -6,7 +6,7 @@ import importlib
from toolbox import update_ui, get_conf
from multiprocessing import Process, Pipe
load_message = "jittorllms尚未加载，加载需要一段时间。注意，取决于`config.py`的配置，jittorllms消耗大量的内存（CPU）或显存（GPU），也许会导致低配计算机卡死 ……"
load_message = "jittorllms尚未加载，加载需要一段时间。注意，请避免混用多种jittor模型，否则可能导致显存溢出而造成卡顿，取决于`config.py`的配置，jittorllms消耗大量的内存（CPU）或显存（GPU），也许会导致低配计算机卡死 ……"
#################################################################################
class GetGLMHandle(Process):
@ -15,6 +15,7 @@ class GetGLMHandle(Process):
self.parent, self.child = Pipe()
self.jittorllms_model = None
self.info = ""
self.local_history = []
self.success = True
self.check_dependency()
self.start()
@ -22,13 +23,14 @@ class GetGLMHandle(Process):
def check_dependency(self):
try:
import jittor
from .jittorllms.models import get_model
import pandas
self.info = "依赖检测通过"
self.success = True
except:
self.info = r"缺少jittorllms的依赖，如果要使用jittorllms，除了基础的pip依赖以外，您还需要运行`pip install -r request_llm/requirements_jittorllms.txt`"+\
r"和`git clone https://gitlink.org.cn/jittor/JittorLLMs.git --depth 1 request_llm/jittorllms`两个指令来安装jittorllms的依赖（在项目根目录运行这两个指令）。"
from toolbox import trimmed_format_exc
self.info = r"缺少jittorllms的依赖，如果要使用jittorllms，除了基础的pip依赖以外，您还需要运行`pip install -r request_llm/requirements_jittorllms.txt -i https://pypi.jittor.org/simple -I`"+\
r"和`git clone https://gitlink.org.cn/jittor/JittorLLMs.git --depth 1 request_llm/jittorllms`两个指令来安装jittorllms的依赖（在项目根目录运行这两个指令）。" +\
r"警告：安装jittorllms依赖后将完全破坏现有的pytorch环境，建议使用docker环境！" + trimmed_format_exc()
self.success = False
def ready(self):
@ -37,6 +39,16 @@ class GetGLMHandle(Process):
def run(self):
# 子进程执行
# 第一次运行，加载参数
def validate_path():
import os, sys
dir_name = os.path.dirname(__file__)
env = os.environ.get("PATH", "")
os.environ["PATH"] = env.replace('/cuda/bin', '/x/bin')
root_dir_assume = os.path.abspath(os.path.dirname(__file__) + '/..')
os.chdir(root_dir_assume + '/request_llm/jittorllms')
sys.path.append(root_dir_assume + '/request_llm/jittorllms')
validate_path() # validate path so you can run from base directory
def load_model():
import types
try:
@ -44,23 +56,37 @@ class GetGLMHandle(Process):
device, = get_conf('LOCAL_MODEL_DEVICE')
from .jittorllms.models import get_model
# availabel_models = ["chatglm", "pangualpha", "llama", "chatrwkv"]
args_dict = {'model': 'chatglm', 'RUN_DEVICE':'cpu'}
args_dict = {'model': 'chatrwkv'}
print('self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))')
self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))
print('done get model')
except:
self.child.send('[Local Message] Call jittorllms fail 不能正常加载jittorllms的参数。')
raise RuntimeError("不能正常加载jittorllms的参数！")
print('load_model')
load_model()
# 进入任务等待状态
print('进入任务等待状态')
while True:
# 进入任务等待状态
kwargs = self.child.recv()
# 收到消息，开始请求
query = kwargs['query']
history = kwargs['history']
# 是否重置
if len(self.local_history) > 0 and len(history)==0:
print('触发重置')
self.jittorllms_model.reset()
self.local_history.append(query)
print('收到消息，开始请求')
try:
for response, history in self.jittorllms_model.run_web_demo(kwargs['query'], kwargs['history']):
for response in self.jittorllms_model.stream_chat(query, history):
print(response)
self.child.send(response)
except:
from toolbox import trimmed_format_exc
print(trimmed_format_exc())
self.child.send('[Local Message] Call jittorllms fail.')
# 请求处理结束，开始下一个循环
self.child.send('[Finish]')
@ -77,32 +103,32 @@ class GetGLMHandle(Process):
break
self.threadLock.release()
global glm_handle
glm_handle = None
global rwkv_glm_handle
rwkv_glm_handle = None
#################################################################################
def predict_no_ui_long_connection(inputs, llm_kwargs, history=[], sys_prompt="", observe_window=[], console_slience=False):
"""
多线程方法
函数的说明请见 request_llm/bridge_all.py
"""
global glm_handle
if glm_handle is None:
glm_handle = GetGLMHandle()
if len(observe_window) >= 1: observe_window[0] = load_message + "\n\n" + glm_handle.info
if not glm_handle.success:
error = glm_handle.info
glm_handle = None
global rwkv_glm_handle
if rwkv_glm_handle is None:
rwkv_glm_handle = GetGLMHandle()
if len(observe_window) >= 1: observe_window[0] = load_message + "\n\n" + rwkv_glm_handle.info
if not rwkv_glm_handle.success:
error = rwkv_glm_handle.info
rwkv_glm_handle = None
raise RuntimeError(error)
# jittorllms 没有 sys_prompt 接口，因此把prompt加入 history
history_feedin = []
history_feedin.append(["What can I do?", sys_prompt])
for i in range(len(history)//2):
history_feedin.append([history[2*i], history[2*i+1]] )
watch_dog_patience = 5 # 看门狗 (watchdog) 的耐心, 设置5秒即可
response = ""
for response in glm_handle.stream_chat(query=inputs, history=history_feedin, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
for response in rwkv_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=sys_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
print(response)
if len(observe_window) >= 1: observe_window[0] = response
if len(observe_window) >= 2:
if (time.time()-observe_window[1]) > watch_dog_patience:
@ -118,13 +144,13 @@ def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_promp
"""
chatbot.append((inputs, ""))
global glm_handle
if glm_handle is None:
glm_handle = GetGLMHandle()
chatbot[-1] = (inputs, load_message + "\n\n" + glm_handle.info)
global rwkv_glm_handle
if rwkv_glm_handle is None:
rwkv_glm_handle = GetGLMHandle()
chatbot[-1] = (inputs, load_message + "\n\n" + rwkv_glm_handle.info)
yield from update_ui(chatbot=chatbot, history=[])
if not glm_handle.success:
glm_handle = None
if not rwkv_glm_handle.success:
rwkv_glm_handle = None
return
if additional_fn is not None:
@ -136,13 +162,12 @@ def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_promp
# 处理历史信息
history_feedin = []
history_feedin.append(["What can I do?", system_prompt] )
for i in range(len(history)//2):
history_feedin.append([history[2*i], history[2*i+1]] )
# 开始接收jittorllms的回复
response = "[Local Message]: 等待jittorllms响应中 ..."
for response in glm_handle.stream_chat(query=inputs, history=history_feedin, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
for response in rwkv_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=system_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
chatbot[-1] = (inputs, response)
yield from update_ui(chatbot=chatbot, history=history)

5
request_llm/requirements_jittorllms.txt
View File

@ -1,4 +1,7 @@
jittor >= 1.3.7.9
jtorch >= 0.1.3
torch
torchvision
torchvision
transformers==4.26.1
pandas
jieba

61
request_llm/test_llms.py
View File

@ -1,6 +1,6 @@
"""
对各个llm模型进行单元测试
"""
# """
# 对各个llm模型进行单元测试
# """
def validate_path():
import os, sys
dir_name = os.path.dirname(__file__)
@ -10,7 +10,9 @@ def validate_path():
validate_path() # validate path so you can run from base directory
from request_llm.bridge_jittorllms import predict_no_ui_long_connection
from request_llm.bridge_jittorllms_rwkv import predict_no_ui_long_connection
# from request_llm.bridge_jittorllms_pangualpha import predict_no_ui_long_connection
# from request_llm.bridge_jittorllms_llama import predict_no_ui_long_connection
llm_kwargs = {
'max_length': 512,
@ -22,5 +24,54 @@ result = predict_no_ui_long_connection(inputs="你好",
llm_kwargs=llm_kwargs,
history=[],
sys_prompt="")
print('final result:', result)
print('result')
result = predict_no_ui_long_connection(inputs="what is a hero?",
llm_kwargs=llm_kwargs,
history=["hello world"],
sys_prompt="")
print('final result:', result)
result = predict_no_ui_long_connection(inputs="如何理解传奇?",
llm_kwargs=llm_kwargs,
history=[],
sys_prompt="")
print('final result:', result)
# # print(result)
# from multiprocessing import Process, Pipe
# class GetGLMHandle(Process):
# def __init__(self):
# super().__init__(daemon=True)
# pass
# def run(self):
# # 子进程执行
# # 第一次运行，加载参数
# def validate_path():
# import os, sys
# dir_name = os.path.dirname(__file__)
# root_dir_assume = os.path.abspath(os.path.dirname(__file__) + '/..')
# os.chdir(root_dir_assume + '/request_llm/jittorllms')
# sys.path.append(root_dir_assume + '/request_llm/jittorllms')
# validate_path() # validate path so you can run from base directory
# jittorllms_model = None
# import types
# try:
# if jittorllms_model is None:
# from models import get_model
# # availabel_models = ["chatglm", "pangualpha", "llama", "chatrwkv"]
# args_dict = {'model': 'chatrwkv'}
# print('self.jittorllms_model = get_model(types.SimpleNamespace(**args_dict))')
# jittorllms_model = get_model(types.SimpleNamespace(**args_dict))
# print('done get model')
# except:
# # self.child.send('[Local Message] Call jittorllms fail 不能正常加载jittorllms的参数。')
# raise RuntimeError("不能正常加载jittorllms的参数！")
# x = GetGLMHandle()
# x.start()
# input()