• Stars
    star
    2,528
  • Rank 18,142 (Top 0.4 %)
  • Language
  • Created almost 2 years ago
  • Updated 3 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

总结Prompt&LLM论文,开源数据&模型,AIGC应用

DecryptPrompt

如果LLM的突然到来让你感到沮丧,不妨读下主目录的Choose Your Weapon Survival Strategies for Depressed AI Academics 持续更新以下内容,Star to keep updated~

目录顺序如下

  1. 国内外,垂直领域大模型
  2. Agent和指令微调等训练框架
  3. 开源指令,预训练,rlhf,对话,agent训练数据梳理
  4. AIGC相关应用
  5. prompt写作指南和5星博客等资源梳理
  6. Prompt和LLM论文细分方向梳理

My blogs & ChatGPT应用

LLMS

模型评测

大模型评估尚未出现北极星指标,整体上全面性有余,泛化性不足,类perplexity的指标还未出现,期待ing ~

榜单 结果
AlpacaEval:LLM-based automatic evaluation 开源模型王者vicuna,openchat, wizardlm
Huggingface Open LLM Leaderboard 只评估开源模型,Falcon夺冠,在Eleuther AI4个评估集上评估的LLM模型榜单,vicuna夺冠
Berkley出品大模型排位赛榜有准中文榜单 Elo评分机制,GPT4自然是稳居第一,GPT4>Claude>GPT3.5>Vicuna>others
CMU开源聊天机器人评测应用 ChatGPT>Vicuna>others;在对话场景中训练可能很重要
Z-Bench中文真格基金评测 国产中文模型的编程可用性还相对较低,大家水平差不太多,两版ChatGLM提升明显
Chain-of-thought评估 GSM8k, MATH等复杂问题排行榜
InfoQ 大模型综合能力评估 面向中文,ChatGPT>文心一言> Claude>星火
ToolBench: 工具调用评估榜单 工具微调模型和ChatGPT进行对比,提供评测脚本
AgentBench: 推理决策评估榜单 清华联合多高校推出不同任务环境,例如购物,家居,操作系统等场景下模型推理决策能力
FlagEval 智源出品主观+客观LLM评分榜单
Bird-Bench 更贴合真实世界应用的超大数据库,需要领域知识的NL2SQL榜单,模型追赶人类尚有时日
kola 以世界知识为核心的评价基准,包括已知的百科知识和未知的近90天网络发布内容,评价知识记忆,理解,应用和创造能力

国外模型

模型链接 模型描述
LLama2 Open Meta带着可商用开源的羊驼2模型来了~
Vicuna Alpaca前成员等开源以LLama13B为基础使用ShareGPT指令微调的模型,提出了用GPT4来评测模型效果
WizardLM 微软新发布13B,登顶AlpacaEval开源模型Top3,使用ChatGPT对指令进行复杂度进化微调LLama2
OpenChat 80k ShareGPT对话微调LLama-2 13B开源模型中的战斗机
Guanaco LLama 7B基座,在alpaca52K数据上加入534K多语言指令数据微调
Google Bard 谷歌bard支持中文了
Claude ChatGPT最大竞争对手Claude也开放申请了,slack中无限试用
LLaMA Meta开源指令微调LLM,规模70 亿到 650 亿不等
MPT MosaicML开源的预训练+指令微调的新模型,可商用,支持84k tokens超长输入
Falcon Falcon由阿联酋技术研究所在超高质量1万亿Token上训练得到1B,7B,40B开源,免费商用!土豪们表示钱什么的格局小了
RedPajama RedPajama项目既开源预训练数据后开源3B,7B的预训练+指令微调模型
koala 使用alpaca,HC3等开源指令集+ ShareGPT等ChatGPT数据微调llama,在榜单上排名较高
ChatLLaMA 基于RLHF微调了LLaMA
Alpaca 斯坦福开源的使用52k数据在7B的LLaMA上微调得到,
Alpaca-lora LORA微调的LLaMA
Dromedary IBM self-aligned model with the LLaMA base
ColossalChat HPC-AI Tech开源的Llama+RLHF微调
MiniGPT4 Vicuna+BLIP2 文本视觉融合
StackLLama LLama使用Stackexchange数据+SFT+RL
Cerebras Cerebras开源了1亿到130亿的7个模型,从预训练数据到参数全开源
PaLM-E 谷歌多模态大模型,540B的PaLM语言模型和22B的ViT视觉模型相结合,得到562B的PaLM-E模型,在机器人应用场景有了新的突破
Dolly-v2 可商用 7b指令微调开源模型在GPT-J-6B上微调
OpenChatKit openai研究员打造GPT-NoX-20B微调+6B审核模型过滤
MetaLM 微软开源的大规模自监督预训练模型
Amazon Titan 亚马逊在aws上增加自家大模型
OPT-IML Meta复刻GPT3,up to 175B, 不过效果并不及GPT3
Bloom BigScience出品,规模最大176B
BloomZ BigScience出品, 基于Bloom微调
Galacia 和Bloom相似,更针对科研领域训练的模型
T0 BigScience出品,3B~11B的在T5进行指令微调的模型
EXLLama Python/C++/CUDA implementation of Llama for use with 4-bit GPTQ weight
LongChat llama-13b使用condensing rotary embedding technique微调的长文本模型
MPT-30B MosaicML开源的在8Ktoken上训练的大模型

国内模型

模型链接 模型描述
ChatGLM2 32K长文本,FlashAttention+Multi-Query Attenion的显存优化,更强推理能力,哈哈不过很多简单问题也硬要COT,中英平行能力似乎略有下降的ChatGLM2,但是免费商用!
ChatGLM 清华开源的、支持中英双语的对话语言模型,使用了代码训练,指令微调和RLHF。chatglm2支持超长文本,可免费商用啦!
LLama2-chinese 没等太久中文预训练微调后的llama2它来了~
YuLan-chat2 高瓴人工智能基于Llama-2中英双语继续预训练+指令微调/对话微调
ziya IDEA研究院在7B/13B llama上继续预训练+SFT+RM+PPO+HFTT+COHFT+RBRS
Baichuan 百川智能开源7B大模型可商用免费
Baichuan2 百川第二代,提供了7B/13B Base和chat的版本
Chinese-LLaMA-Alpaca 哈工大中文指令微调的LLaMA
Moss 为复旦正名!开源了预训练,指令微调的全部数据和模型。可商用
Aquila 智源开源7B大模型可商用免费
Qwen-7B 阿里开源,可商用,通义千文7B模型
XVERSE 据说中文超越llama2的元象开源模型13B模型
BiLLa LLama词表扩充预训练+预训练和任务1比1混合SFT+指令样本SFT三阶段训练
Phoenix 港中文开源凤凰和奇美拉LLM,Bloom基座,40+语言支持
Wombat-7B 达摩院开源无需强化学习使用RRHF对齐的语言模型, alpaca基座
TigerBot 虎博开源了7B 180B的模型以及预训练和微调语料
Luotuo 中文指令微调的LLaMA,和ChatGLM
OpenBuddy Llama 多语言对话微调模型
Chinese Vincuna LLama 7B基座,使用Belle+Guanaco数据训练
Linly Llama 7B基座,使用belle+guanaco+pclue+firefly+CSL+newscommentary等7个指令微调数据集训练
Firefly 中文2.6B模型,提升模型中文写作,古文能力,待开源全部训练代码,当前只有模型
Baize 使用100k self-chat对话数据微调的LLama
BELLE 使用ChatGPT生成数据对开源模型进行中文优化
Chatyuan chatgpt出来后最早的国内开源对话模型,T5架构是下面PromptCLUE的衍生模型
PromptCLUE 多任务Prompt语言模型
PLUG 阿里达摩院发布的大模型,提交申请会给下载链接
CPM2.0 智源发布CPM2.0
GLM 清华发布的中英双语130B预训练模型
BayLing 基于LLama7B/13B,增强的语言对齐的英语/中文大语言模型
文心一言 已经拿到邀请码并试用,虽然人格化程度显著低,但效果上并没有很拉胯,国产YYDS!不过商业化霸王条款确实不少
通义千问 阿里系LLM开放申请
星火 科大讯飞星火,数学是真的厉害

垂直领域模型&进展

领域 模型链接 模型描述
医疗 MedGPT 医联发布的
医疗 MedPalm Google在Faln-PaLM的基础上通过多种类型的医疗QA数据进行prompt-tuning指令微调得到,同时构建了MultiMedQA
医疗 ChatDoctor 110K真实医患对话样本+5KChatGPT生成数据进行指令微调
医疗 Huatuo Med-ChatGLM 医学知识图谱和chatgpt构建中文医学指令数据集+医学文献和chatgpt构建多轮问答数据
医疗 Chinese-vicuna-med Chinese-vicuna在cMedQA2数据上微调
医疗 OpenBioMed 清华AIR开源轻量版BioMedGPT, 知识图谱&20+生物研究领域多模态预训练模型
医疗 DoctorGLM ChatDoctor+MedDialog+CMD 多轮对话+单轮指令样本微调GLM
医疗 MedicalGPT-zh 自建的医学数据库ChatGPT生成QA+16个情境下SELF构建情景对话
医疗 PMC-LLaMA 医疗论文微调Llama
医疗 PULSE Bloom微调+继续预训练
医疗 NHS-LLM Chatgpt生成的医疗问答,对话,微调模型
医疗 神农医疗大模型 以中医知识图谱的实体为中心生成的中医知识指令数据集11w+,微调LLama-7B
医疗 岐黄问道大模型 3个子模型构成,已确诊疾病的临床治疗模型+基于症状的临床诊疗模型+中医养生条理模型,看起来是要ToB落地
医疗 MeChat 心理咨询领域,通过chatgpt改写多轮对话56k
医疗 SoulChat 心理咨询领域中文长文本指令与多轮共情对话数据联合指令微调 ChatGLM-6B
医疗 MindChat MindChat-Baichuan-13B,Qwen-7B,MindChat-InternLM-7B使用不同基座在模型安全,共情,人类价值观对其上进行了强化
医疗 DISC-MedLLM 疾病知识图谱构建QA对+QA对转化成单论对话+真实世界数据重构+人类偏好数据筛选,SFT微调baichuan
法律 LawGPT-zh 利用ChatGPT清洗CrimeKgAssitant数据集得到52k单轮问答+我们根据中华人民共和国法律手册上最核心的9k法律条文,利用ChatGPT联想生成具体的情景问答+知识问答使用ChatGPT基于文本构建QA对
法律 LawGPT 基于llama+扩充词表二次预训练+基于法律条款构建QA指令微调
法律 Lawyer Llama 法律指令微调数据集:咨询+法律考试+对话进行指令微调
法律 LexiLaw 法律指令微调数据集:问答+书籍概念解释,法条内容进行指令微调
法律 ChatLaw 北大推出的法律大模型,应用形式很新颖类似频道内流一切功能皆融合在对话形式内
法律 录问模型 在baichuan基础上40G二次预训练+100K指令微调,在知识库构建上采用了Emb+意图+关键词联想结合的方案
金融 FinChat.io 使用最新的财务数据,电话会议记录,季度和年度报告,投资书籍等进行训练
金融 OpenGPT 领域LLM指令样本生成+微调框架
金融 乾元BigBang金融2亿模型 金融领域预训练+任务微调
金融 度小满千亿金融大模型 在Bloom-176B的基础上进行金融+中文预训练和微调
金融 bondGPT GPT4在细分债券市场的应用开放申请中
金融 IndexGPT JPMorgan在研的生成式投资顾问
金融 恒生LightGPT 金融领域继续预训练+插件化设计
金融 知彼阿尔法 企查查商查大模型
金融 AlphaBox 熵简科技发布大模型金融应用,多文档问答+会议转录+文档编辑
金融 曹植 达观发布金融大模型融合data2text等金融任务,赋能报告写作
金融 聚宝盆 基于 LLaMA 系基模型经过中文金融知识指令精调/指令微调(Instruct-tuning) 的微调模型
金融 PIXIU 整理了多个金融任务数据集加入了时间序列数据进行指令微调
金融 ChatFund 韭圈儿发布的第一个基金大模型,看起来是做了多任务指令微调,和APP已有的数据功能进行了全方位的打通,从选基,到持仓分析等等
编程 Starcoder 80种编程语言+Issue+Commit训练得到的编程大模型
编程 ChatSQL 基于ChatGLM实现NL2sql
编程 codegeex 13B预训练+微调多语言变成大模型
编程 codegeex2 Chatglm2的基础上CodeGeeX2-6B 进一步经过了 600B 代码数据预训练
编程 stabelcode 560B token多语言预训练+ 120,000 个 Alpaca指令对齐
编程 SQLCoder 在StarCoder的基础上微调15B超越gpt3.5
数学 MathGPT 是好未来自主研发的,面向全球数学爱好者和科研机构,以解题和讲题算法为核心的大模型。
交通 TransGPT LLama-7B+34.6万领域预训练+5.8万条领域指令对话微调(来自文档问答)
科技 Mozi 红睡衣预训练+论文QA数据集 + ChatGPT扩充科研对话数据
天文 StarGLM 天文知识指令微调,项目进行中后期考虑天文二次预训练+KG
写作 阅文-网文大模型介绍 签约作者内测中,主打的内容为打斗场景,剧情切换,环境描写,人设,世界观等辅助片段的生成
写作 MediaGPT LLama-7B扩充词表+指令微调,指令来自国内媒体专家给出的在新闻创作上的80个子任务
电商 EcomGPT 电商领域任务指令微调大模型,指令样本250万,基座模型是Bloomz

Tool and Library

指令微调,预训练,rlhf框架

工具描述 链接
LoRA:Low-Rank指令微调方案 https://github.com/tloen/alpaca-lora
peft:parameter-efficient prompt tunnging工具集 https://github.com/huggingface/peft
RL4LMs:AllenAI的RL工具 https://github.com/allenai/RL4LMs
trl:基于Transformer的强化训练框架 https://github.com/lvwerra/trl
trlx:分布式训练trl https://github.com/CarperAI/trlx
北大开源河狸项目可复现RLHF,支持多数LLM,提供RLHF数据 https://github.com/PKU-Alignment/safe-rlhf
RL4LMs:AllenAI的RL工具 https://github.com/allenai/RL4LMs
LMFlow:港科大实验室开源的大模型微调框架,支持以上多数开源模型的指令微调和RLHF https://github.com/OptimalScale/LMFlow
hugNLP:基于Huggingface开发继承Prompt技术,预训练和是指输入等多种方案 https://github.com/wjn1996/HugNLP
Deepspeed:针对RL训练和推理的整合优化 https://github.com/microsoft/DeepSpeed
Uerpy:预训练框架支持lm,mlm,unilm等 https://github.com/dbiir/UER-py
TecentPretrain: Uerpy的重构版本支持llama预训练 https://github.com/Tencent/TencentPretrain/tree/main
lamini: 整合指令数据生成,SFT,RLHF的工具库 https://github.com/lamini-ai/lamini/
Chain-of-thought-hub:模型推理能力评估平台 https://github.com/FranxYao/chain-of-thought-hub
FlexGen:LLM推理 CPU Offload计算架构 https://github.com/FMInference/FlexGen
VLLM:超高速推理框架Vicuna,Arena背后的无名英雄,比HF快24倍 https://github.com/vllm-project/vllm
EasyEdit:浙大开源支持多种模型,多种方案的模型知识精准编辑器 https://github.com/zjunlp/EasyEdit
FlexFlow:模型部署推理框架 https://github.com/flexflow/FlexFlow
Medusa:针对采样解码的推理加速框架,可以和其他策略结合 https://github.com/FasterDecoding/Medusa

LLM Agent工具

工具描述 链接
langchain:LLM Agent框架 https://github.com/hwchase17/langchain
llama index:LLM Agent框架 https://github.com/jerryjliu/llama_index
semantic-kernel:整合大模型和编程语言的SDK https://github.com/microsoft/semantic-kernel
BMTTools: 清华出品多工具调用开源库,提供微调数据和评估ToolBench https://github.com/OpenBMB/BMTools
BabyAGI:自执行LLM Agent https://github.com/yoheinakajima/babyagi
AutoGPT:自执行LLM Agent https://github.com/Torantulino/Auto-GPT
MetaGPT: 覆盖软件公司全生命流程,例如产品经理等各个职业的AutoGPT https://github.com/geekan/MetaGPT
ResearchGPT: 论文写作领域的AutoGPT,融合论文拆解+网络爬虫 https://github.com/assafelovic/gpt-researcher
MiniAGI:自执行LLM Agent https://github.com/muellerberndt/mini-agi
AL Legion: 自执行LLM Agent https://github.com/eumemic/ai-legion
AgentVerse:多模型交互环境 https://github.com/OpenBMB/AgentVerse
AgentSims: 给定一个社会环境,评估LLM作为智能体的预定任务目标完成能力的沙盒环境
GPTRPG:RPG环境 AI Agent游戏化 https://github.com/dzoba/gptrpg
GPTEngineer:自动工具构建和代码生成 https://github.com/AntonOsika/gpt-engineer
Jarvis: 大模型调用小模型框架,给小模型一个未来! https://github.com/search?q=jarvis
LLM-ToolMaker:让LLM自己制造Agent https://github.com/FMInference/FlexGen
Gorilla: LLM调用大量API https://github.com/ShishirPatil/gorilla
wenda:闻达小模型整合搜索用于知识融入 https://github.com/l15y/wenda
WorkGPT:类似AutoGPT https://github.com/team-openpm/workgpt
Deep-KE:基于LLM对数据进行智能解析实现知识抽取 https://github.com/zjunlp/DeepKE
Vectra:平台化的LLM Agent搭建方案,从索引构建,内容召回排序,到事实检查的LLM生成 https://vectara.com/tour-vectara/
Alexandria: 从Arix论文开始把整个互联网变成向量索引,可以免费下载 https://alex.macrocosm.so/download
RapidAPI: 统一这个世界的所有API,最大API Hub,有调用成功率,latency等,是真爱! https://rapidapi.com/hub
Data-Copilot:时间序列等结构化数据分析领域的Agent解决方案 https://github.com/zwq2018/Data-Copilot
DB-GPT: 以数据库为基础的GPT实验项目,使用本地化的GPT大模型与您的数据和环境进行交互 https://db-gpt.readthedocs.io/projects/db-gpt-docs-zh-cn/zh_CN/latest/index.html
guardrails:降低模型幻觉的python框架,promp模板+validation+修正 https://github.com/shreyar/guardrails
guidance:微软新开源框架,同样是降低模型幻觉的框架,prompt+chain的升级版加入逐步生成和思维链路 https://github.com/guidance-ai/guidance
Ragas: 评估检索增强LLM效果的框架,基于大模型prompt评估事实性,召回相关性,召回内容质量,回答相关性等 https://github.com/explodinggradients/ragas#fire-quickstart
langflow:把langchain等agent组件做成了可拖拽式的UI https://github.com/logspace-ai/langflow

Training Data

数据类型 数据描述 数据链接
指令微调 self-instruct,GPT3自动生成&过滤得到指令集 https://github.com/yizhongw/self-instruct
指令微调 Standford Alpaca:52K text-davinci-003生成的self-instruct指令数据集 https://github.com/tatsu-lab/stanford_alpaca
指令微调 GPT4-for-LLM 中文+英文+对比指令 https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM
指令微调 GPTTeacher更多样的通用指令,角色扮演和代码指令 https://github.com/teknium1/GPTeacher/tree/main
指令微调 中文翻译Alpaca还有一些其他指令数据集 https://github.com/hikariming/alpaca_chinese_dataset https://github.com/carbonz0/alpaca-chinese-dataset
指令微调 alpaca指令GPT4生成,和以上几版对比显著质量更高,回复更长 https://github.com/Instruction-Tuning-with-GPT-4/GPT-4-LLM/tree/main
指令微调 Guanaco数据:对Alphca指令重写后以不同语言生成总共534K,有对话和非对话类型,还有补充的QA生成样本 https://huggingface.co/datasets/JosephusCheung/GuanacoDataset
指令微调 OIG中文指令包括翻译alpaca+natural+unnatural,多轮对话,考试,leetcode指令 https://github.com/BAAI-Zlab/COIG
指令微调 Vicuna训练使用的样本,用API获取了sharegpt上用户和chatgpt对话历史,部分网友整理到了HF https://github.com/domeccleston/sharegpt https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/tree/main
指令微调 HC3指令数据中英文,包括金融,开放QA,百科,DBQA,医学等包含人工回复 https://huggingface.co/datasets/Hello-SimpleAI/HC3-Chinese/tree/main
指令微调 MOSS开源的SFT数据包含使用plugin的对话数据 https://huggingface.co/datasets/Hello-SimpleAI/HC3-Chinese/tree/main
指令微调 InstructWild数据:用四处爬取的chatgpt指令作为种子self-instruct扩充生成,中英双语 https://github.com/XueFuzhao/InstructionWild/tree/main/data
指令微调 BELLE100万指令数据,参考Alpaca用ChatGPT生成,有数学,多轮对话,校色对话等等 https://github.com/LianjiaTech/BELLE
指令微调 PromptCLUE多任务提示数据集:模板构建,只包含标准NLP任务 https://github.com/CLUEbenchmark/pCLUE
指令微调 TK-Instruct微调用的指令数据集, 全人工标注1600+NLP任务 https://instructions.apps.allenai.org/
指令微调 T0微调用的指令数据集(P3) https://huggingface.co/datasets/bigscience/P3
指令微调 p3衍生的46种多语言数据集(xmtf) https://github.com/bigscience-workshop/xmtf
指令微调 Unnatural Instruction使用GPT3生成后改写得到240k https://github.com/orhonovich/unnatural-instructions
指令微调 alpaca COT对多个数据源进行了清理并统一格式放到的了HF, 重点是人工整理的COT数据 https://github.com/PhoebusSi/Alpaca-CoT
指令微调 人工编写包含23种常见的中文NLP任务的指令数据,中文写作方向 https://github.com/yangjianxin1/Firefly
指令微调 Amazon COT指令样本包括各类QA,bigbench,math等 https://github.com/amazon-science/auto-cot
指令微调 CSL包含 396,209 篇中文核心期刊论文元信息 (标题、摘要、关键词、学科、门类)可做预训练可构建NLP指令任务 https://github.com/ydli-ai/CSL
指令微调 alpaca code 20K代码指令数据 https://github.com/sahil280114/codealpaca#data-release
指令微调 GPT4Tools 71K GPT4指令样本 https://github.com/StevenGrove/GPT4Tools
指令微调 GPT4指令+角色扮演+代码指令 https://github.com/teknium1/GPTeacher
指令微调 Mol-Instructions 2043K 分子+蛋白质+生物分子文本指令,覆盖分子设计、蛋白质功能预测、蛋白质设计等任务 https://github.com/zjunlp/Mol-Instructions
数学 腾讯人工智能实验室发布网上爬取的数学问题APE210k https://github.com/Chenny0808/ape210k
数学 猿辅导 AI Lab开源小学应用题Math23K https://github.com/SCNU203/Math23k/tree/main
数学 grade school math把OpenAI的高中数学题有改造成指令样本有2-8步推理过程 https://huggingface.co/datasets/qwedsacf/grade-school-math-instructions
数学 数学问答数据集有推理过程和多项选择 https://huggingface.co/datasets/math_qa/viewer/default/test?row=2
数学 AMC竞赛数学题 https://huggingface.co/datasets/competition_math
数学 线性代数等纯数学计算题 https://huggingface.co/datasets/math_dataset
代码 APPS从不同的开放访问编码网站Codeforces、Kattis 等收集的问题 https://opendatalab.org.cn/APPS
代码 Lyra代码由带有嵌入式 SQL 的 Python 代码组成,经过仔细注释的数据库操作程序,配有中文评论和英文评论。 https://opendatalab.org.cn/Lyra
代码 Conala来自StackOverflow问题,手动注释3k,英文 https://opendatalab.org.cn/CoNaLa/download
代码 code-alpaca ChatGPT生成20K代码指令样本 https://github.com/sahil280114/codealpaca.git
代码 32K, 四种不同类型、不同难度的代码相关中文对话数据,有大模型生成, https://github.com/zxx000728/CodeGPT
对话 LAION 策划的开放指令通用数据集中手动选择的组件子集 已开源40M 3万个,100M在路上 https://github.com/LAION-AI/Open-Instruction-Generalist
对话 Baize基于Chat GPT构建的self-chat数据 https://github.com/project-baize/baize-chatbot/tree/main/data
对话 FaceBook开源BlenderBot训练对话数据~6K https://huggingface.co/datasets/blended_skill_talk
对话 AllenAI开源38.5万个对话高质量数据集SODA https://realtoxicityprompts.apps.allenai.org/
对话 InstructDial在单一对话任务类型上进行指令微调 https://github.com/prakharguptaz/Instructdial
对话 Ultra Chat 两个独立的 ChatGPT Turbo API 进行对话,从而生成多轮对话数据 https://github.com/thunlp/UltraChat
对话 Awesome Open-domain Dialogue Models提供多个开放域对话数据 https://github.com/cingtiye/Awesome-Open-domain-Dialogue-Models#%E4%B8%AD%E6%96%87%E5%BC%80%E6%94%BE%E5%9F%9F%E5%AF%B9%E8%AF%9D%E6%95%B0%E6%8D%AE%E9%9B%86
对话 Salesforce开源超全DialogStudio https://github.com/salesforce/DialogStudio
对话 基于事实Reference的多轮问答中文数据,已开源5万,之后会开源更多 https://github.com/sufengniu/RefGPT
RLFH 北大河狸开源RLHF数据集10K,1M需要申请 https://huggingface.co/datasets/PKU-Alignment/PKU-SafeRLHF-10K
RLHF Anthropic hh-rlhf数据集 https://huggingface.co/datasets/Anthropic/hh-rlhf
RLHF Stack-exchange上问题对应多个答案,每个答案有打分 https://huggingface.co/datasets/HuggingFaceH4/stack-exchange-preferences/tree/main
RLHF Facebook Bot Adversarial Dialogues数据集5K https://github.com/facebookresearch/ParlAI
RLHF AllenAI Real Toxicity prompts https://github.com/facebookresearch/ParlAI
RLHF OpenAssistant Conversations 160K消息,13500人工生成, 英文为主 https://huggingface.co/datasets/OpenAssistant/oasst1
RLHF 知乎问答偏好数据集 https://huggingface.co/datasets/liyucheng/zhihu_rlhf_3k
RLHF hh-rlhf中文翻译偏好数据 https://huggingface.co/datasets/liswei/rm-static-zhTW
评估集 BigBench(Beyond the Imitation Game Benchmark) https://github.com/google/BIG-bench
评估集 Complex QA:用于ChatGPT的评测指令集 https://github.com/tan92hl/Complex-Question-Answering-Evaluation-of-ChatGPT
评估集 Langchain开源评估数据集 https://huggingface.co/LangChainDatasets
评估集 2010-2022年全国高考卷的题目 https://github.com/OpenLMLab/GAOKAO-Bench
评估集 中文通用大模型综合性评测基准SuperCLUE https://github.com/CLUEbenchmark/SuperCLUE
英文预训练 RedPajama开源的复刻llama的预训练数据集,1.21万亿Token https://github.com/togethercomputer/RedPajama-Data
英文预训练 Cerebras基于RedPajama进行清洗去重后得到的高质量数据集, 6270亿Token https://huggingface.co/datasets/cerebras/SlimPajama-627B/tree/main/train
英文预训练 Pile 22个高质量数据集混合的预训练数据集800G,全量开放下载 https://pile.eleuther.ai/
通用预训练 UER整理CLUECorpusSmall+News Commentary中英 https://github.com/dbiir/UER-py/wiki/%E9%A2%84%E8%AE%AD%E7%BB%83%E6%95%B0%E6%8D%AE
中文预训练 智源人工智能开源的wudao 200G预训练数据 https://github.com/BAAI-WuDao/WuDaoMM
中文预训练 里屋社区发起开源力量收集中文互联网语料集MNBVC目标是对标ChatGPT的40T https://github.com/esbatmop/MNBVC
中文预训练 复旦开源15万中文图书下载和抽取方案 https://github.com/FudanNLPLAB/CBook-150K
中文预训练 书生万卷数据集来自公开网页多模态数据集,包括文本,图文和视频,其中文本1T,图文150G https://opendatalab.org.cn/OpenDataLab/WanJuan1_dot_0
领域预训练 首个中文科学文献数据集CSL,也有多种NLP任务数据 https://github.com/ydli-ai/CSL
平行语料 news-commentary中英平行语料,用于中英间知识迁移 https://data.statmt.org/news-commentary/v15/training/
多源数据集整合 opendatalab整合了预训练阶段的多个数据源 https://opendatalab.org.cn/?industry=9821&source=JUU3JTlGJUE1JUU0JUI5JThF
Tool-搜索增强 webCPM开源的和搜索工具进行交互问答的数据集,包括网页抽取式摘要,多事实内容回答等人工标注数据 https://github.com/thunlp/WebCPM
Tool-多工具 BmTools开源的多工具调用指令数据集 https://github.com/OpenBMB/BMTools
NL2SQL DB-GPT-Hub梳理了多源text-to-sql数据集 https://github.com/eosphoros-ai/DB-GPT-Hub

AIGC

  • NexusGPT: : AutoGPT可以出来工作了,第一个全AI Freelance平台
  • cognosys: 全网最火的web端AutoGPT,不过咋说呢试用了下感觉下巴要笑掉了,不剧透去试试你就知道
  • godmode:可以进行人为每一步交互的的AutoGPT
  • agentgpt: 基础版AutoGPT
  • do Anything: AutoGPT Like的to Do List生成器
  • ChatMind: chatgpt生成思维导图,模板很丰富,泛化性也不错,已经被XMind收购了~
  • New Bing:需要连外网否则会重定向到bing中国,需要申请waitlist
  • Perplexity.ai: 同样需要科学上网,感觉比Bing做的更好的接入ChatGPT的神奇搜索引擎,在Bing之外还加入了相关推荐和追问
  • BingGPT: NewBing开源桌面客户端,可以将聊天记录导出
  • DocsGPT: 把ChatGPT开放域问答转化成封闭域问答的通用方案,试用垂类领域问答场景,可以试用定制的ChatBot
  • langchain-ChatGLM: 基于ChatGLM的本地知识问答,和上面的DocsGPT相似,不过可以本地部署⭐
  • ChatPDF: 国内的ChatPDF, 上传pdf后,会给出文章的Top5可能问题,然后对话式从文档中进行问答和检索,10s读3万字
  • ChatDoc:ChatPDF升级版,增加了表格类解析,和完善的索引引用加跳转加对应文章内容高亮,哈哈我准备自己整一个
  • ChatPaper: 根据输入关键词,自动在arxiv上下载最新的论文,并对论文进行摘要总结,可以在huggingface上试用!
  • OpenRead: 面向论文写作,阅读场景,可以帮助生成文献综述,以及提供和NotionAI相似的智能Markdown用于写作
  • researchgpt: 和ChatPDF类似,支持arivx论文下载,加载后对话式获取论文重点
  • BriefGPT: 日更Arxiv论文,并对论文进行摘要,关键词抽取,帮助研究者了解最新动态, UI不错哟
  • ChatGPT-academic: 又是一个基于gradio实现的paper润色,摘要等功能打包的实现
  • feishu-chatgpt: 飞书chatgpt,和365copilot相似也是多组件集成, 有点全!
  • AI Topiah: 聆心智能AI角色聊天,和路飞唠了两句,多少有点中二之魂在燃烧
  • chatbase: 情感角色聊天,还没尝试
  • Vana: virtual DNA, 通过聊天创建虚拟自己!概念很炫
  • WriteSonic:AI写作,支持对话和定向创作如广告文案,商品描述, 支持Web检索是亮点,支持中文
  • copy.ai: WriteSonic竞品,亮点是像论文引用一样每句话都有对应网站链接,可以一键复制到右边的创作Markdown,超级好用!
  • NotionAI:智能Markdown,适用真相!在创作中用command调用AI辅助润色,扩写,检索内容,给创意idea
  • Jasper: 同上,全是竞品哈哈
  • copy.down: 中文的营销文案生成,只能定向创作,支持关键词到文案的生成
  • ChatExcel: 指令控制excel计算,对熟悉excel的有些鸡肋,对不熟悉的有点用
  • ChatPPT: 使用ChatGPT进行PPT制作
  • BibiGPT: Bilibli视频内容一键总结,多模态文档
  • Copilot: 要付费哟
  • Fauxpilot: copilot本地开源替代
  • CodeGex: 国内替代品,还没试过
  • Codeium: Copilot替代品,有免费版本支持各种plugin
  • sql translate: text2sql,利用 OpenAI 的 API 实现的一个很简单的工具,sql到文字,文字到sql
  • ai2sql: text2sql老牌公司,相比sqltranslate功能更全面,支持SQL 语法检查、格式化和生成公式
  • chat2query: text2sql 相比以上两位支持更自然的文本指令,以及更复杂的数据分析类的sql生成
  • OuterBase: text2sql 设计风格很吸睛!电子表格结合mysql和dashboard,更适合数据分析宝宝
  • Wolverine: 代码自我debug的python脚本
  • dreamstudio.ai: 开创者,Stable Difussion, 有试用quota
  • midjourney: 开创者,艺术风格为主
  • Dall.E: 三巨头这就凑齐了
  • ControlNet: 为绘画创作加持可控性
  • GFPGAN: 照片修复
  • Visual ChatGPT: 微软发布图像ChatGPT,对话方式进行图像生成编辑,问答
  • gemo.ai: 多模态聊天机器人,包括文本,图像,视频生成
  • storybird: 根据提示词生成故事绘本,还可以售卖

Resources

教程类

书籍博客类

Papers

paper List

综述

  • A Survey of Large Language Models
  • Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing ⭐
  • Paradigm Shift in Natural Language Processing
  • Pre-Trained Models: Past, Present and Future
  • What Language Model Architecture and Pretraining objects work best for zero shot generalization ⭐
  • Towards Reasoning in Large Language Models: A Survey
  • Reasoning with Language Model Prompting: A Survey ⭐
  • An Overview on Language Models: Recent Developments and Outlook ⭐
  • A Survey of Large Language Models[6.29更新版]
  • Unifying Large Language Models and Knowledge Graphs: A Roadmap
  • Augmented Language Models: a Survey ⭐
  • Domain Specialization as the Key to Make Large Language Models Disruptive: A Comprehensive Survey
  • Challenges and Applications of Large Language Models

大模型能力探究

  • In Context Learning
    • LARGER LANGUAGE MODELS DO IN-CONTEXT LEARNING DIFFERENTLY
    • How does in-context learning work? A framework for understanding the differences from traditional supervised learning
    • Why can GPT learn in-context? Language Model Secretly Perform Gradient Descent as Meta-Optimizers ⭐
    • Rethinking the Role of Demonstrations What Makes incontext learning work? ⭐
    • Trained Transformers Learn Linear Models In-Context
  • 涌现能力
    • Sparks of Artificial General Intelligence: Early experiments with GPT-4
    • Emerging Ability of Large Language Models ⭐
  • 能力评估
    • IS CHATGPT A GENERAL-PURPOSE NATURAL LANGUAGE PROCESSING TASK SOLVER?
    • Can Large Language Models Infer Causation from Correlation?
    • Holistic Evaluation of Language Model
    • Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond
    • Theory of Mind May Have Spontaneously Emerged in Large Language Models
    • Beyond The Imitation Game: Quantifying And Extrapolating The Capabilities Of Language Models
    • Do Models Explain Themselves? Counterfactual Simulatability of Natural Language Explanations
    • Demystifying GPT Self-Repair for Code Generation
    • Evidence of Meaning in Language Models Trained on Programs
    • Can Explanations Be Useful for Calibrating Black Box Models
    • On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspective
    • Language acquisition: do children and language models follow similar learning stages?

Prompt Tunning范式

  • Tunning Free Prompt
    • GPT2: Language Models are Unsupervised Multitask Learners
    • GPT3: Language Models are Few-Shot Learners ⭐
    • LAMA: Language Models as Knowledge Bases?
    • AutoPrompt: Eliciting Knowledge from Language Models
  • Fix-Prompt LM Tunning
    • T5: Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
    • PET-TC(a): Exploiting Cloze Questions for Few Shot Text Classification and Natural Language Inference ⭐
    • PET-TC(b): PETSGLUE It’s Not Just Size That Matters Small Language Models are also few-shot learners
    • GenPET: Few-Shot Text Generation with Natural Language Instructions
    • LM-BFF: Making Pre-trained Language Models Better Few-shot Learners ⭐
    • ADEPT: Improving and Simplifying Pattern Exploiting Training
  • Fix-LM Prompt Tunning
    • Prefix-tuning: Optimizing continuous prompts for generation
    • Prompt-tunning: The power of scale for parameter-efficient prompt tuning ⭐
    • P-tunning: GPT Understands Too ⭐
    • WARP: Word-level Adversarial ReProgramming
  • LM + Prompt Tunning
    • P-tunning v2: Prompt Tuning Can Be Comparable to Fine-tunning Universally Across Scales and Tasks
    • PTR: Prompt Tuning with Rules for Text Classification
    • PADA: Example-based Prompt Learning for on-the-fly Adaptation to Unseen Domains
  • Fix-LM Adapter Tunning
    • LORA: LOW-RANK ADAPTATION OF LARGE LANGUAGE MODELS ⭐
    • LST: Ladder Side-Tuning for Parameter and Memory Efficient Transfer Learning
    • Parameter-Efficient Transfer Learning for NLP
    • INTRINSIC DIMENSIONALITY EXPLAINS THE EFFECTIVENESS OF LANGUAGE MODEL FINE-TUNING

主流LLMS

  • GLM-130B: AN OPEN BILINGUAL PRE-TRAINED MODEL
  • LLaMA: Open and Efficient Foundation Language Models
  • PaLM: Scaling Language Modeling with Pathways
  • PaLM 2 Technical Report
  • GPT-4 Technical Report
  • Backpack Language Models
  • Llama 2: Open Foundation and Fine-Tuned Chat Models

指令微调&对齐 (instruction_tunning)

  • 经典方案
    • Flan: FINETUNED LANGUAGE MODELS ARE ZERO-SHOT LEARNERS ⭐
    • Flan-T5: Scaling Instruction-Finetuned Language Models
    • ExT5: Towards Extreme Multi-Task Scaling for Transfer Learning
    • Instruct-GPT: Training language models to follow instructions with human feedback ⭐
    • T0: MULTITASK PROMPTED TRAINING ENABLES ZERO-SHOT TASK GENERALIZATION
    • Natural Instructions: Cross-Task Generalization via Natural Language Crowdsourcing Instructions
    • Tk-INSTRUCT: SUPER-NATURALINSTRUCTIONS: Generalization via Declarative Instructions on 1600+ NLP Tasks
    • ZeroPrompt: Scaling Prompt-Based Pretraining to 1,000 Tasks Improves Zero-shot Generalization
    • Unnatural Instructions: Tuning Language Models with (Almost) No Human Labor
    • INSTRUCTEVAL Towards Holistic Evaluation of Instrucion-Tuned Large Language Models
  • 更少,质量更高、更多样的指令数据带来质变
    • LIMA: Less Is More for Alignment ⭐
    • Maybe Only 0.5% Data is Needed: A Preliminary Exploration of Low Training Data Instruction Tuning
    • Textbooks Are All You Need ⭐
    • AlpaGasus: Training A Better Alpaca with Fewer Data
    • InstructionGPT-4: A 200-Instruction Paradigm for Fine-Tuning MiniGPT-4
    • Instruction Mining: High-Quality Instruction Data Selection for Large Language Models
    • Visual Instruction Tuning with Polite Flamingo
  • 新对齐/微调方案
    • WizardLM: Empowering Large Language Models to Follow Complex Instructions
    • Becoming self-instruct: introducing early stopping criteria for minimal instruct tuning
    • Self-Alignment with Instruction Backtranslation ⭐
    • Mixture-of-Experts Meets Instruction Tuning:A Winning Combination for Large Language Models
    • Goat: Fine-tuned LLaMA Outperforms GPT-4 on Arithmetic Tasks
    • PROMPT2MODEL: Generating Deployable Models from Natural Language Instructions
    • OpinionGPT: Modelling Explicit Biases in Instruction-Tuned LLMs
  • 微调经验/实验报告
    • BELLE: Exploring the Impact of Instruction Data Scaling on Large Language Models: An Empirical Study on Real-World Use Cases
    • Baize: Baize: An Open-Source Chat Model with Parameter-Efficient Tuning on Self-Chat Data
    • A Comparative Study between Full-Parameter and LoRA-based Fine-Tuning on Chinese Instruction Data for Large LM
    • Exploring ChatGPT’s Ability to Rank Content: A Preliminary Study on Consistency with Human Preferences
    • Towards Better Instruction Following Language Models for Chinese: Investigating the Impact of Training Data and Evaluation

对话模型

  • LaMDA: Language Models for Dialog Applications
  • Sparrow: Improving alignment of dialogue agents via targeted human judgements ⭐
  • BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage
  • How NOT To Evaluate Your Dialogue System: An Empirical Study of Unsupervised Evaluation Metrics for Dialogue Response Generation
  • DialogStudio: Towards Richest and Most Diverse Unified Dataset Collection for Conversational AI
  • Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
  • DiagGPT: An LLM-based Chatbot with Automatic Topic Management for Task-Oriented Dialogue

思维链 (prompt_chain_of_thought)

  • 基础&进阶用法
    • [zero-shot-COT] Large Language Models are Zero-Shot Reasoners ⭐
    • [few-shot COT] Chain of Thought Prompting Elicits Reasoning in Large Language Models ⭐
    • SELF-CONSISTENCY IMPROVES CHAIN OF THOUGHT REASONING IN LANGUAGE MODELS
    • LEAST-TO-MOST PROMPTING ENABLES COMPLEX REASONING IN LARGE LANGUAGE MODELS ⭐
    • Tree of Thoughts: Deliberate Problem Solving with Large Language Models
    • Plan-and-Solve Prompting: Improving Zero-Shot Chain-of-Thought Reasoning by Large Language Models
    • Decomposed Prompting A MODULAR APPROACH FOR Solving Complex Tasks
    • Successive Prompting for Decomposing Complex Questions
    • Verify-and-Edit: A Knowledge-Enhanced Chain-of-Thought Framework
    • Beyond Chain-of-Thought, Effective Graph-of-Thought Reasoning in Large Language Models
    • Tree-of-Mixed-Thought: Combining Fast and Slow Thinking for Multi-hop Visual Reasoning
    • LAMBADA: Backward Chaining for Automated Reasoning in Natural Language
  • 分领域COT [Math, Code, Tabular, QA]
    • Solving Quantitative Reasoning Problems with Language Models
    • SHOW YOUR WORK: SCRATCHPADS FOR INTERMEDIATE COMPUTATION WITH LANGUAGE MODELS
    • Solving math word problems with processand outcome-based feedback
    • CodeRL: Mastering Code Generation through Pretrained Models and Deep Reinforcement Learning
    • T-SciQ: Teaching Multimodal Chain-of-Thought Reasoning via Large Language Model Signals for Science Question Answering
    • LEARNING PERFORMANCE-IMPROVING CODE EDITS
    • Large Language Models are Versatile Decomposers: Decompose Evidence and Questions for Table-based Reasoning
    • Tab-CoT: Zero-shot Tabular Chain of Thought
  • 原理分析
    • Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters ⭐
    • TEXT AND PATTERNS: FOR EFFECTIVE CHAIN OF THOUGHT IT TAKES TWO TO TANGO
    • Towards Revealing the Mystery behind Chain of Thought: a Theoretical Perspective
    • Large Language Models Can Be Easily Distracted by Irrelevant Context
  • 小模型COT蒸馏
    • Specializing Smaller Language Models towards Multi-Step Reasoning ⭐
    • Teaching Small Language Models to Reason
    • Large Language Models are Reasoning Teachers
    • Distilling Reasoning Capabilities into Smaller Language Models
    • The CoT Collection: Improving Zero-shot and Few-shot Learning of Language Models via Chain-of-Thought Fine-Tuning
  • COT样本自动构建/选择
    • STaR: Self-Taught Reasoner Bootstrapping ReasoningWith Reasoning
    • AutoCOT:AUTOMATIC CHAIN OF THOUGHT PROMPTING IN LARGE LANGUAGE MODELS
    • Large Language Models Can Self-Improve
    • Active Prompting with Chain-of-Thought for Large Language Models
    • COMPLEXITY-BASED PROMPTING FOR MULTI-STEP REASONING
  • others
    • OlaGPT Empowering LLMs With Human-like Problem-Solving abilities
    • Challenging BIG-Bench tasks and whether chain-of-thought can solve them
    • Large Language Models are Better Reasoners with Self-Verification
    • ThoughtSource A central hub for large language model reasoning data
    • Two Failures of Self-Consistency in the Multi-Step Reasoning of LLMs

RLHF

  • Deepmind
    • Teaching language models to support answers with verified quotes
    • sparrow, Improving alignment of dialogue agents via targetd human judgements ⭐
  • openai
    • PPO: Proximal Policy Optimization Algorithms ⭐
    • Deep Reinforcement Learning for Human Preference
    • Fine-Tuning Language Models from Human Preferences
    • learning to summarize from human feedback
    • InstructGPT: Training language models to follow instructions with human feedback ⭐
    • Scaling Laws for Reward Model Over optimization ⭐
  • Anthropic
    • A General Language Assistant as a Laboratory for Alignmen
    • Red Teaming Language Models to Reduce Harms Methods,Scaling Behaviors and Lessons Learned
    • Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback ⭐
    • Constitutional AI Harmlessness from AI Feedback ⭐
    • Pretraining Language Models with Human Preferences
    • The Capacity for Moral Self-Correction in Large Language Models
  • AllenAI, RL4LM:IS REINFORCEMENT LEARNING (NOT) FOR NATURAL LANGUAGE PROCESSING BENCHMARKS
  • 改良方案
    • RRHF: Rank Responses to Align Language Models with Human Feedback without tears
    • PRM:Let's verify step by step
    • Chain of Hindsight Aligns Language Models with Feedback
    • AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedback
    • Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
    • RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignment

LLM Agent 让模型使用工具 (llm_agent)

  • 基于prompt通用方案
    • ReAct: SYNERGIZING REASONING AND ACTING IN LANGUAGE MODELS ⭐
    • Self-ask: MEASURING AND NARROWING THE COMPOSITIONALITY GAP IN LANGUAGE MODELS ⭐
    • MRKL SystemsA modular, neuro-symbolic architecture that combines large language models, external knowledge sources and discrete reasoning
    • PAL: Program-aided Language Models
    • ART: Automatic multi-step reasoning and tool-use for large language models
    • ReWOO: Decoupling Reasoning from Observations for Efficient Augmented Language Models ⭐
    • Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions
    • Chameleon: Plug-and-Play Compositional Reasoning with Large Language Models ⭐
    • Faithful Chain-of-Thought Reasoning
    • Reflexion: Language Agents with Verbal Reinforcement Learning ⭐
    • Search-in-the-Chain: Towards Accurate, Credible and Traceable Large Language Models for Knowledge-intensive Tasks
    • Verify-and-Edit: A Knowledge-Enhanced Chain-of-Thought Framework
    • RestGPT: Connecting Large Language Models with Real-World RESTful APIs
  • 基于微调通用方案
    • TALM: Tool Augmented Language Models
    • Toolformer: Language Models Can Teach Themselves to Use Tools ⭐
    • Tool Learning with Foundation Models
    • Tool Maker:Large Language Models as Tool Maker
    • TaskMatrix.AI: Completing Tasks by Connecting Foundation Models with Millions of APIs
  • 检索增强方案
    • WebGPT:Browser-assisted question-answering with human feedback
    • WebGLM: Towards An Efficient Web-Enhanced Question Answering System with Human Preferences
    • WebCPM: Interactive Web Search for Chinese Long-form Question Answering ⭐
    • REPLUG: Retrieval-Augmented Black-Box Language Models
    • Query Rewriting for Retrieval-Augmented Large Language Models
    • RETA-LLM: A Retrieval-Augmented Large Language Model Toolkit
    • Atlas: Few-shot Learning with Retrieval Augmented Language Models
    • RRAML: Reinforced Retrieval Augmented Machine Learning
  • 调用模型方案
    • HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in HuggingFace
    • Gorilla:Large Language Model Connected with Massive APIs ⭐
    • OpenAGI: When LLM Meets Domain Experts
  • 垂直领域
    • WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents
    • ToolkenGPT: Augmenting Frozen Language Models with Massive Tools via Tool Embeddings
    • ChemCrow Augmenting large language models with chemistry tools
    • Data-Copilot: Bridging Billions of Data and Humans with Autonomous Workflow
    • GeneGPT: Augmenting Large Language Models with Domain Tools for Improved Access to Biomedical Information
    • PointLLM: Empowering Large Language Models to Understand Point Clouds
  • 评估
    • Evaluating Verifiability in Generative Search Engines
    • Mind2Web: Towards a Generalist Agent for the Web
    • Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions
    • API-Bank: A Benchmark for Tool-Augmented LLMs
    • ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs
    • FACTSCORE: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
  • 智能体
    • Generative Agents: Interactive Simulacra of Human Behavior
    • CAMEL: Communicative Agents for "Mind" Exploration of Large Scale Language Model Society
    • LLM+P: Empowering Large Language Models with Optimal Planning Proficiency
  • 其他
    • Inference with Reference: Lossless Acceleration of Large Language Models
    • RecallM: An Architecture for Temporal Context Understanding and Question Answering

指令数据生成 (instruction_data_gen)

  • APE: LARGE LANGUAGE MODELS ARE HUMAN-LEVEL PROMPT ENGINEERS ⭐
  • SELF-INSTRUCT: Aligning Language Model with Self Generated Instructions ⭐
  • iPrompt: Explaining Data Patterns in Natural Language via Interpretable Autoprompting
  • Flipped Learning: Guess the Instruction! Flipped Learning Makes Language Models Stronger Zero-Shot Learners
  • Fairness-guided Few-shot Prompting for Large Language Models
  • Instruction induction: From few examples to natural language task descriptions.
  • Baize An Open-Source Chat Model with Parameter-Efficient Tuning on self-Chat Data
  • SELF-QA Unsupervised Knowledge Guided alignment.
  • GPT Self-Supervision for a Better Data Annotator
  • The Flan Collection Designing Data and Methods
  • Self-Consuming Generative Models Go MAD
  • InstructEval: Systematic Evaluation of Instruction Selection Methods
  • Overwriting Pretrained Bias with Finetuning Data

预训练数据(pretrain_data)

  • DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining
  • The Pile: An 800GB Dataset of Diverse Text for Language Modeling
  • CCNet: Extracting High Quality Monolingual Datasets fromWeb Crawl Data
  • WanJuan: A Comprehensive Multimodal Dataset for Advancing English and Chinese Large Models
  • CLUECorpus2020: A Large-scale Chinese Corpus for Pre-training Language Model

领域模型 (domain_llms)

  • MedGPT: Medical Concept Prediction from Clinical Narratives
  • BioGPT:Generative Pre-trained Transformer for Biomedical Text Generation and Mining
  • Galactia:A Large Language Model for Science
  • PubMed GPT: A Domain-specific large language model for biomedical text ⭐
  • BloombergGPT: A Large Language Model for Finance
  • ChatDoctor:Medical Chat Model Fine-tuned on LLaMA Model using Medical Domain Knowledge
  • Med-PaLM:Large Language Models Encode Clinical Knowledge[V1,V2] ⭐
  • Augmented Large Language Models with Parametric Knowledge Guiding
  • XuanYuan 2.0: A Large Chinese Financial Chat Model with Hundreds of Billions Parameters
  • ChatLaw Open-Source Legal Large Language Model ⭐
  • MediaGPT : A Large Language Model For Chinese Media
  • SMILE: Single-turn to Multi-turn Inclusive Language Expansion via ChatGPT for Mental Health Support
  • KITLM: Domain-Specific Knowledge InTegration into Language Models for Question Answering
  • FinVis-GPT: A Multimodal Large Language Model for Financial Chart Analysis
  • EcomGPT: Instruction-tuning Large Language Models with Chain-of-Task Tasks for E-commerce

LLM超长文本处理 (long_input)

NL2SQL

  • 大模型方案
    • DIN-SQL: Decomposed In-Context Learning of Text-to-SQL with Self-Correction ⭐
    • C3: Zero-shot Text-to-SQL with ChatGPT ⭐
    • SQL-PALM: IMPROVED LARGE LANGUAGE MODEL ADAPTATION FOR TEXT-TO-SQL
    • BIRD Can LLM Already Serve as A Database Interface? A BIg Bench for Large-Scale Database Grounded Text-to-SQL ⭐
    • A Case-Based Reasoning Framework for Adaptive Prompting in Cross-Domain Text-to-SQL
    • ChatDB: AUGMENTING LLMS WITH DATABASES AS THEIR SYMBOLIC MEMORY
    • A comprehensive evaluation of ChatGPT’s zero-shot Text-to-SQL capability
    • Few-shot Text-to-SQL Translation using Structure and Content Prompt Learning
  • Domain Knowledge Intensive
    • Towards Knowledge-Intensive Text-to-SQL Semantic Parsing with Formulaic Knowledge
    • Bridging the Generalization Gap in Text-to-SQL Parsing with Schema Expansion
    • Towards Robustness of Text-to-SQL Models against Synonym Substitution
    • FinQA: A Dataset of Numerical Reasoning over Financial Data
  • others
    • RESDSQL: Decoupling Schema Linking and Skeleton Parsing for Text-to-SQL
    • MIGA: A Unified Multi-task Generation Framework for Conversational Text-to-SQL

降低模型幻觉 (reliability)

  • Survey of Hallucination in Natural Language Generation
  • Trusting Your Evidence: Hallucinate Less with Context-aware Decoding ⭐
  • SELF-REFINE:ITERATIVE REFINEMENT WITH SELF-FEEDBACK ⭐
  • PROMPTING GPT-3 TO BE RELIABLE
  • Enhancing Self-Consistency and Performance of Pre-Trained Language Models through Natural Language Inference
  • On the Advance of Making Language Models Better Reasoners
  • Progressive-Hint Prompting Improves Reasoning in Large Language Models
  • ASK ME ANYTHING: A SIMPLE STRATEGY FOR PROMPTING LANGUAGE MODELS ⭐
  • Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
  • Reflexion: an autonomous agent with dynamic memory and self-reflection
  • Self-consistency for open-ended generations
  • Check Your Facts and Try Again: Improving Large Language Models with External Knowledge and Automated Feedback
  • Factuality Enhanced Language Models for Open-Ended Text Generation
  • Adaptive Chameleon or Stubborn Sloth: Unraveling the Behavior of Large Language Models in Knowledge Clashes
  • Rethinking with Retrieval: Faithful Large Language Model Inference
  • KoLA: Carefully Benchmarking World Knowledge of Large Language Models
  • RefGPT: Reference → Truthful & Customized Dialogues Generation by GPTs and for GPTs

推理速度优化(fast_inference)

  • Fast Transformer Decoding: One Write-Head is All You Need
  • Fast Inference from Transformers via Speculative Decoding
  • GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints
  • Skeleton-of-Thought: Large Language Models Can Do Parallel Decoding
  • SkipDecode: Autoregressive Skip Decoding with Batching and Caching for Efficient LLM Inference

模型知识编辑黑科技(model_edit)

  • ROME:Locating and Editing Factual Associations in GPT
  • Transformer Feed-Forward Layers Are Key-Value Memories
  • MEMIT: Mass-Editing Memory in a Transformer
  • MEND:Fast Model Editing at Scale
  • Editing Large Language Models: Problems, Methods, and Opportunities

Other Prompt Engineer(prompt_engineer)

  • Calibrate Before Use: Improving Few-Shot Performance of Language Models
  • In-Context Instruction Learning
  • LEARNING PERFORMANCE-IMPROVING CODE EDITS
  • Boosting Theory-of-Mind Performance in Large Language Models via Prompting
  • Generated Knowledge Prompting for Commonsense Reasoning
  • RECITATION-AUGMENTED LANGUAGE MODELS
  • kNN PROMPTING: BEYOND-CONTEXT LEARNING WITH CALIBRATION-FREE NEAREST NEIGHBOR INFERENCE
  • EmotionPrompt: Leveraging Psychology for Large Language Models Enhancement via Emotional Stimulus
  • Causality-aware Concept Extraction based on Knowledge-guided Prompting

Multimodal

  • InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning
  • Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models
  • PaLM-E: An Embodied Multimodal Language Model
  • LLava Visual Instruction Tuning
  • MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models
  • TabLLM: Few-shot Classification of Tabular Data with Large Language Models
  • BLIVA: A Simple Multimodal LLM for Better Handling of Text-Rich Visual Questions
  • mPLUG-Owl : Modularization Empowers Large Language Models with Multimodality
  • LVLM eHub: A Comprehensive Evaluation Benchmark for Large VisionLanguage Models

Others

  • Learnware: Small Models Do Big
  • The economic potential of generative AI
  • A PhD Student’s Perspective on Research in NLP in the Era of Very Large Language Models