int/gpt_academic
1
0
派生 0
镜像自地址 https://github.com/binary-husky/gpt_academic.git 已同步 2025-12-06 14:36:48 +00:00
代码 工单 软件包 项目 版本发布 百科 动态

Merge branch 'frontier' into production

浏览代码
这个提交包含在：
binary-husky
2023-11-11 23:51:10 +08:00
父节点 9023aa6732 eeb70e966c
当前提交 cad541d8d7
共有 40 个文件被更改，包括 318 次插入 和 252 次删除
下载 Patch 文件 下载 Diff 文件 展开所有文件 折叠所有文件

9
request_llms/bridge_all.py
查看文件

@@ -242,6 +242,13 @@ for model in AVAIL_LLM_MODELS:
mi.update({"endpoint": api2d_endpoint})
model_info.update({model: mi})
# -=-=-=-=-=-=- azure 对齐支持 -=-=-=-=-=-=-
for model in AVAIL_LLM_MODELS:
if model.startswith('azure-') and (model.replace('azure-','') in model_info.keys()):
mi = model_info[model.replace('azure-','')]
mi.update({"endpoint": azure_endpoint})
model_info.update({model: mi})
# -=-=-=-=-=-=- 以下部分是新加入的模型，可能附带额外依赖 -=-=-=-=-=-=-
if "claude-1-100k" in AVAIL_LLM_MODELS or "claude-2" in AVAIL_LLM_MODELS:
from .bridge_claude import predict_no_ui_long_connection as claude_noui
@@ -564,7 +571,7 @@ 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, llm_kwargs, history, sys_prompt, observe_window=[], console_slience=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。
inputs：

3
request_llms/bridge_chatglm.py
查看文件

@@ -4,14 +4,13 @@ cmd_to_install = "`pip install -r request_llms/requirements_chatglm.txt`"
from transformers import AutoModel, AutoTokenizer
from toolbox import get_conf, ProxyNetworkActivate
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns, SingletonLocalLLM
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 Local Model
# ------------------------------------------------------------------------------------------------------------------------
@SingletonLocalLLM
class GetGLM2Handle(LocalLLMHandle):
def load_model_info(self):

3
request_llms/bridge_chatglm3.py
查看文件

@@ -4,14 +4,13 @@ cmd_to_install = "`pip install -r request_llms/requirements_chatglm.txt`"
from transformers import AutoModel, AutoTokenizer
from toolbox import get_conf, ProxyNetworkActivate
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns, SingletonLocalLLM
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 Local Model
# ------------------------------------------------------------------------------------------------------------------------
@SingletonLocalLLM
class GetGLM3Handle(LocalLLMHandle):
def load_model_info(self):

3
request_llms/bridge_chatglmonnx.py
查看文件

@@ -8,7 +8,7 @@ import threading
import importlib
from toolbox import update_ui, get_conf
from multiprocessing import Process, Pipe
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns, SingletonLocalLLM
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
from .chatglmoonx import ChatGLMModel, chat_template
@@ -17,7 +17,6 @@ from .chatglmoonx import ChatGLMModel, chat_template
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 Local Model
# ------------------------------------------------------------------------------------------------------------------------
@SingletonLocalLLM
class GetONNXGLMHandle(LocalLLMHandle):
def load_model_info(self):

3
request_llms/bridge_chatgpt.py
查看文件

@@ -7,8 +7,7 @@
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：高级实验性功能模块调用，不会实时显示在界面上，参数简单，可以多线程并行，方便实现复杂的功能逻辑
3. predict_no_ui_long_connection：在实验过程中发现调用predict_no_ui处理长文档时，和openai的连接容易断掉，这个函数用stream的方式解决这个问题，同样支持多线程
2. predict_no_ui_long_connection：支持多线程
"""
import json

3
request_llms/bridge_chatgpt_website.py
查看文件

@@ -7,8 +7,7 @@
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：高级实验性功能模块调用，不会实时显示在界面上，参数简单，可以多线程并行，方便实现复杂的功能逻辑
3. predict_no_ui_long_connection：在实验过程中发现调用predict_no_ui处理长文档时，和openai的连接容易断掉，这个函数用stream的方式解决这个问题，同样支持多线程
2. predict_no_ui_long_connection：支持多线程
"""
import json

2
request_llms/bridge_claude.py
查看文件

@@ -7,7 +7,7 @@
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：在实验过程中发现调用predict_no_ui处理长文档时，和openai的连接容易断掉，这个函数用stream的方式解决这个问题，同样支持多线程
2. predict_no_ui_long_connection：支持多线程
"""
import os

23
request_llms/bridge_internlm.py
查看文件

@@ -5,9 +5,9 @@ from transformers import AutoModel, AutoTokenizer
import time
import threading
import importlib
from toolbox import update_ui, get_conf
from toolbox import update_ui, get_conf, ProxyNetworkActivate
from multiprocessing import Process, Pipe
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns, SingletonLocalLLM
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
# ------------------------------------------------------------------------------------------------------------------------
@@ -34,7 +34,6 @@ def combine_history(prompt, hist):
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 Local Model
# ------------------------------------------------------------------------------------------------------------------------
@SingletonLocalLLM
class GetInternlmHandle(LocalLLMHandle):
def load_model_info(self):
@@ -53,14 +52,15 @@ class GetInternlmHandle(LocalLLMHandle):
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = get_conf('LOCAL_MODEL_DEVICE')
if self._model is None:
tokenizer = AutoTokenizer.from_pretrained("internlm/internlm-chat-7b", trust_remote_code=True)
if device=='cpu':
model = AutoModelForCausalLM.from_pretrained("internlm/internlm-chat-7b", trust_remote_code=True).to(torch.bfloat16)
else:
model = AutoModelForCausalLM.from_pretrained("internlm/internlm-chat-7b", trust_remote_code=True).to(torch.bfloat16).cuda()
with ProxyNetworkActivate('Download_LLM'):
if self._model is None:
tokenizer = AutoTokenizer.from_pretrained("internlm/internlm-chat-7b", trust_remote_code=True)
if device=='cpu':
model = AutoModelForCausalLM.from_pretrained("internlm/internlm-chat-7b", trust_remote_code=True).to(torch.bfloat16)
else:
model = AutoModelForCausalLM.from_pretrained("internlm/internlm-chat-7b", trust_remote_code=True).to(torch.bfloat16).cuda()
model = model.eval()
model = model.eval()
return model, tokenizer
def llm_stream_generator(self, **kwargs):
@@ -94,8 +94,9 @@ class GetInternlmHandle(LocalLLMHandle):
inputs = tokenizer([prompt], padding=True, return_tensors="pt")
input_length = len(inputs["input_ids"][0])
device = get_conf('LOCAL_MODEL_DEVICE')
for k, v in inputs.items():
inputs[k] = v.cuda()
inputs[k] = v.to(device)
input_ids = inputs["input_ids"]
batch_size, input_ids_seq_length = input_ids.shape[0], input_ids.shape[-1]
if generation_config is None:

3
request_llms/bridge_llama2.py
查看文件

@@ -5,14 +5,13 @@ cmd_to_install = "`pip install -r request_llms/requirements_chatglm.txt`"
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from toolbox import update_ui, get_conf, ProxyNetworkActivate
from multiprocessing import Process, Pipe
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns, SingletonLocalLLM
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
from threading import Thread
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 Local Model
# ------------------------------------------------------------------------------------------------------------------------
@SingletonLocalLLM
class GetONNXGLMHandle(LocalLLMHandle):
def load_model_info(self):

19
request_llms/bridge_qwen.py
查看文件

@@ -6,16 +6,15 @@ from transformers import AutoModel, AutoTokenizer
import time
import threading
import importlib
from toolbox import update_ui, get_conf
from toolbox import update_ui, get_conf, ProxyNetworkActivate
from multiprocessing import Process, Pipe
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns, SingletonLocalLLM
from .local_llm_class import LocalLLMHandle, get_local_llm_predict_fns
# ------------------------------------------------------------------------------------------------------------------------
# 🔌💻 Local Model
# ------------------------------------------------------------------------------------------------------------------------
@SingletonLocalLLM
class GetONNXGLMHandle(LocalLLMHandle):
def load_model_info(self):
@@ -30,13 +29,13 @@ class GetONNXGLMHandle(LocalLLMHandle):
import platform
from modelscope import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
model_id = 'qwen/Qwen-7B-Chat'
revision = 'v1.0.1'
self._tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision, trust_remote_code=True)
# use fp16
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", revision=revision, trust_remote_code=True, fp16=True).eval()
model.generation_config = GenerationConfig.from_pretrained(model_id, trust_remote_code=True) # 可指定不同的生成长度、top_p等相关超参
self._model = model
with ProxyNetworkActivate('Download_LLM'):
model_id = 'qwen/Qwen-7B-Chat'
self._tokenizer = AutoTokenizer.from_pretrained('Qwen/Qwen-7B-Chat', trust_remote_code=True, resume_download=True)
# use fp16
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True, fp16=True).eval()
model.generation_config = GenerationConfig.from_pretrained(model_id, trust_remote_code=True) # 可指定不同的生成长度、top_p等相关超参
self._model = model
return self._model, self._tokenizer

36
request_llms/local_llm_class.py
查看文件

@@ -1,6 +1,6 @@
import time
import threading
from toolbox import update_ui
from toolbox import update_ui, Singleton
from multiprocessing import Process, Pipe
from contextlib import redirect_stdout
from request_llms.queued_pipe import create_queue_pipe
@@ -26,23 +26,20 @@ class ThreadLock(object):
def __exit__(self, type, value, traceback):
self.release()
def SingletonLocalLLM(cls):
"""
Singleton Decroator for LocalLLMHandle
"""
_instance = {}
@Singleton
class GetSingletonHandle():
def __init__(self):
self.llm_model_already_running = {}
def _singleton(*args, **kargs):
if cls not in _instance:
_instance[cls] = cls(*args, **kargs)
return _instance[cls]
elif _instance[cls].corrupted:
_instance[cls] = cls(*args, **kargs)
return _instance[cls]
def get_llm_model_instance(self, cls, *args, **kargs):
if cls not in self.llm_model_already_running:
self.llm_model_already_running[cls] = cls(*args, **kargs)
return self.llm_model_already_running[cls]
elif self.llm_model_already_running[cls].corrupted:
self.llm_model_already_running[cls] = cls(*args, **kargs)
return self.llm_model_already_running[cls]
else:
return _instance[cls]
return _singleton
return self.llm_model_already_running[cls]
def reset_tqdm_output():
import sys, tqdm
@@ -76,7 +73,6 @@ class LocalLLMHandle(Process):
self.parent_state, self.child_state = create_queue_pipe()
# allow redirect_stdout
self.std_tag = "[Subprocess Message] "
self.child.write = lambda x: self.child.send(self.std_tag + x)
self.running = True
self._model = None
self._tokenizer = None
@@ -137,6 +133,8 @@ class LocalLLMHandle(Process):
def run(self):
# 🏃‍♂️🏃‍♂️🏃‍♂️ run in child process
# 第一次运行，加载参数
self.child.flush = lambda *args: None
self.child.write = lambda x: self.child.send(self.std_tag + x)
reset_tqdm_output()
self.set_state("`尝试加载模型`")
try:
@@ -220,7 +218,7 @@ def get_local_llm_predict_fns(LLMSingletonClass, model_name, history_format='cla
"""
refer to request_llms/bridge_all.py
"""
_llm_handle = LLMSingletonClass()
_llm_handle = GetSingletonHandle().get_llm_model_instance(LLMSingletonClass)
if len(observe_window) >= 1:
observe_window[0] = load_message + "\n\n" + _llm_handle.get_state()
if not _llm_handle.running:
@@ -268,7 +266,7 @@ def get_local_llm_predict_fns(LLMSingletonClass, model_name, history_format='cla
"""
chatbot.append((inputs, ""))
_llm_handle = LLMSingletonClass()
_llm_handle = GetSingletonHandle().get_llm_model_instance(LLMSingletonClass)
chatbot[-1] = (inputs, load_message + "\n\n" + _llm_handle.get_state())
yield from update_ui(chatbot=chatbot, history=[])
if not _llm_handle.running: