Giters

jianzhnie / awesome-open-chatgpt

Open efforts to implement ChatGPT-like models and beyond.

Home Page:https://jianzhnie.github.io/machine-learning-wiki/#/deep-rl/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

chatgptgpt4instruct-gptrlhfllama

Awesome Open Chatgpt

ChatGPT is GPT-3.5 finetuned with RLHF (Reinforcement Learning with Human Feedback) for human instruction and chat.

Alternatives are projects featuring different instruct finetuned language models for chat. Projects are not counted if they are:

  • Alternative frontend projects which simply call OpenAI's APIs.
  • Using language models which are not finetuned for human instruction or chat.

Tags:

  • Bare: only source code, no data, no model's weight, no chat system
  • Standard: yes data, yes model's weight, bare chat via API
  • Full: full yes data, yes model's weight, fancy chat system including TUI and GUI
  • Complicated: semi open source, not really open source, based on closed model, etc...

Other revelant lists:

Table of Contents

Open source ChatGPT list

list of open source works to implement ChatGPT-like models.

institution model language base model Tuning dataset main feature
Meta LLaMA en - togethercomputer/RedPajama-Data-1T LLaMA-13B outperforms GPT-3(175B) and LLaMA-65B is competitive to PaLM-540M.
Base model for most follow-up works.
@ggerganov llama.cpp en LLaMA NA c/cpp implement of llama and some other models, using quantization.
Stanford Alpaca en LLaMA-7B Alpaca use 52K instruction-following data generated by Self-Instructt techniques to fine-utne 7B LLaMA,
the resulting model, Alpaca, behaves similarly to the text-davinci-003 model on the Self-Instruct instruction-following evaluation suite.
Alpaca has inspired many follow-up models.
LianJia BELLE en/zh BLOOMZ-7B1-mt 1.5M中文数据集 maybe the first Chinese model to follow Alpaca.
Tsinghua ChatGLM-6B en/zh GLM NA well-known Chinese model, in chat mode, and can run on single GPU.
Databricks Dolly en GPT-J 6B databricks/databricks-dolly-15k use Alpaca data to fine-tune a 2-year-old model: GPT-J, which exhibits surprisingly high quality
instruction following behavior not characteristic of the foundation model on which it is based.
@tloen Alpaca-LoRA en LLaMA-7B Alpaca trained within hours on a single RTX 4090,
reproducing the Stanford Alpaca results using low-rank adaptation (LoRA),
and can run on a Raspberry pi.
ColossalAI ColossalChat en/zh LLaMA-7B InstructWild Data provides a unified large language model framework, including:
Supervised datasets collection
Supervised instructions fine-tuning
Reward model training
RLHF
Quantization inference
Fast model deploying
Perfectly integrated with the Hugging Face ecosystem
Shanghai AI Lab LLaMA-Adapter en LLaMA-7B NA Fine-tuning LLaMA to follow instructions within 1 Hour and 1.2M Parameters
PhoebusSi Alpaca-CoT en/zh LLaMA
ChatGLM
BLOOM		 extend CoT data to Alpaca to boost its reasoning ability.
aims to build an instruction finetuning (IFT) platform with extensive instruction collection (especially the CoT datasets)
and a unified interface for various large language models.
AetherCortex Llama-X en LLaMA Open Academic Research on Improving LLaMA to SOTA LLM
Together OpenChatKit en GPT-NeoX-20B laion/OIG OpenChatKit provides a powerful, open-source base to create both specialized and general purpose chatbots for various applications.
The kit includes an instruction-tuned language models, a moderation model, and an extensible retrieval system for including
up-to-date responses from custom repositories.
nomic-ai GPT4All en LLaMA trained on a massive collection of clean assistant data including code, stories and dialogue
@ymcui Chinese-LLaMA-Alpaca en/zh LLaMA-7B/13B expand the Chinese vocabulary based on the original LLaMA and use Chinese data for secondary pre-training,
further enhancing Chinese basic semantic understanding. Additionally, the project uses Chinese instruction data
for fine-tuning on the basis of the Chinese LLaMA, significantly improving the model's understanding and execution of instructions.
UC Berkley
Stanford
CMU		 Vicuna en LLaMA-13B Impressing GPT-4 with 90% ChatGPT Quality
@NouamaneTazi bloomz.cpp en/zh BLOOM C++ implementation for BLOOM inference.
HKUST LMFlow en/zh LLaMA
Galatica
GPT-2
...		 An extensible, convenient, and efficient toolbox for finetuning large machine learning models, designed to be user-friendly,
speedy and reliable, and accessible to the entire community.
Cerebras Systems Cerebras-GPT en - Pretrained LLM, GPT-3 like, Commercially available, efficiently trained on theAndromeda AI supercomputer,
trained in accordance with Chinchilla scaling laws (20 tokens per model parameter) which is compute-optimal.
UT Southwestern/
UIUC/OSU/HDU		 ChatDoctor en LLaMA ChatDoctor Dataset Maybe the first domain-specific chat model tuned on LLaMA.
LAION-AI Open-Assistant en/zh Llama OpenAssistant/oasst1 Open Assistant is a project meant to give everyone access to a great chat based large language model.
project-baize baize-chatbot En/zh llama baize-chatbot Baize is an open-source chat model trained with LoRA. It uses 100k dialogs generated by letting ChatGPT chat with itself. We also use Alpaca's data to improve its performance. We have released 7B, 13B and 30B models. Please refer to the paper for more details.

The template

Append the new project at the end of file

## [{owner}/{project-name}]{https://github.com/link/to/project}

Description goes here

Tags: Bare/Standard/Full/Complicated

The list

lucidrains/PaLM-rlhf-pytorch

Implementation of RLHF (Reinforcement Learning with Human Feedback) on top of the PaLM architecture. Basically ChatGPT but with PaLM

Tags: Bare stars: ⭐⭐⭐

LAION-AI/Open-Assistant

Open Assistant is a truly open-source project, which means giving everyone access to top chat-based large language models. It aims to create a revolution in innovation in language by enabling people to interact with third-party systems, retrieve information dynamically, and create new applications using language.

Related links:

Tags: Full Stars: ⭐⭐⭐⭐⭐

ColossalAI/Chat

ColossalChat implement LLM with RLHF, powered by the Colossal-AI project.

Tags: full Stars: ⭐⭐⭐⭐⭐

nebuly-ai/nebullvm

由于 LLaMA 大模型系列没有使用 RLHF 方法，因此初创公司 Nebuly AI 开源了 RLHF 版 LLaMA（ChatLLaMA）的训练方法。它的训练过程类似 ChatGPT，该项目允许基于预训练的 LLaMA 模型构建 ChatGPT 形式的服务。与 ChatGPT 相比，LLaMA 架构更小，但训练过程和单 GPU 推理速度更快，成本更低；该库还支持所有的 LLaMA 模型架构（7B、13B、33B、65B），因此用户可以根据训练时间和推理性能偏好对模型进行微调。

ChatLLAMA is a chatbot powered by LLaMA, a large language model finetuned with RLHF.

Tags: Full Stars: ⭐⭐⭐⭐⭐

Stability-AI/StableLM

This repository contains Stability AI's ongoing development of the StableLM series of language models and will be continuously updated with new checkpoints.

Related links:

Tags: Full

DeepSpeed-Chat

DeepSpeed-Chat: Easy, Fast and Affordable RLHF Training of ChatGPT-like Models at All Scales. A fast, affordable, scalable and open system framework for enabling end-to-end Reinforcement Learning Human Feedback (RLHF) training experience to generate high-quality ChatGPT-style models at all scales.

Related links:

togethercomputer/OpenChatKit

OpenChatKit provides a powerful, open-source base to create both specialized and general purpose chatbots for various applications.

Related links:

Tags: Full Stars: ⭐⭐⭐⭐⭐

tatsu-lab/stanford_alpaca

This is the repo for the Stanford Alpaca project, which aims to build and share an instruction-following LLaMA model.

Tags: Complicated Stars: ⭐⭐⭐

Resources:

lm-sys/FastChat

An open platform for training, serving, and evaluating large language model based chatbots. Vicuna is an open-source chatbot with 13B parameters trained by fine-tuning LLaMA on user conversations data collected from ShareGPT.com, a community site users can share their ChatGPT conversations. Based on evaluations done, the model has a more than 90% quality rate comparable to OpenAI's ChatGPT and Google's Bard, which makes this model one of the top opensourced models when looking at feature parity to ChatGPT.

Resources:

THUDM/ChatGLM-6B

ChatGLM-6B is an open bilingual language model based on General Language Model (GLM) framework, with 6.2 billion parameters. With the quantization technique, users can deploy locally on consumer-grade graphics cards (only 6GB of GPU memory is required at the INT4 quantization level).

Related links:

Tags: Full

Baize 白泽

Baize exhibits impressive performance in multi-turn dialogues thanks to its guardrails that help mitigate potential risks. It has achieved this through a high-quality multi-turn chat corpus, which was developed by leveraging ChatGPT to facilitate conversations with itself.

Baize code source, model, and dataset are released under a non-commercial (research purposes) license.

databrickslabs/dolly

Databricks’ Dolly, a large language model trained on the Databricks Machine Learning Platform. Script to fine tune GPT-J 6B model on the Alpaca dataset. Insightful if you want to fine tune LLMs.

Related links:

Tags: Standard

Stars: ⭐⭐⭐⭐

h2oai/h2ogpt

h2oGPT - The world's best open source GPT

  • Open-source repository with fully permissive, commercially usable code, data and models
  • Code for preparing large open-source datasets as instruction datasets for fine-tuning of large language models (LLMs), including prompt engineering
  • Code for fine-tuning large language models (currently up to 20B parameters) on commodity hardware and enterprise GPU servers (single or multi node)
  • Code to run a chatbot on a GPU server, with shareable end-point with Python client API
  • Code to evaluate and compare the performance of fine-tuned LLMs

Related links:

Tags: Full

Panda: 海外中文开源大语言模型

Panda: 海外中文开源大语言模型，基于 Llama-7B, -13B, -33B, -65B 进行中文领域上的持续预训练。

数据

模型数据现阶段均采用开源的公开中英文语料数据集：

中文 instruction-tuning

Notes

  1. 对于除维基百科和新闻语料外的其他语料，用 Conditional Generation 的方式优化，即 instruction 部分与输入部分不计算损失，只计算输出部分的损失。除 COIG 外的语料中的 instruction 为固定模板。
  2. 一开始我们将以上所有语料混合在一起进行训练，但发现最终的模型在 instruction following 方面的能力并不好，因此我们决定单独在 COIG 数据集上进行指令微调，并得到最终模型。推测原因可能是 COIG 在整体训练数据中的占比过小，可选的解决方案是对 COIG 加大采样的概率。

英文 instruction-tuning

为了提升模型的基础能力，我们选择使用 FLAN Collection 进行训练。由于 FLAN collection 语料规模过于庞大，我们按比例抽取了 7M 的语料用于训练，且最终性能仍远落后于 FLAN-T5-3B，因此目前我们决定暂时停止该方向的训练，并思考其他可能的构建较小的同时具有较强基础能力的语言模型的方向。

clue-ai/ChatYuan

ChatYuan: Large Language Model for Dialogue in Chinese and English (The repos are mostly in Chinese)

Related links:

Tags: Full

nomic-ai/gpt4all

GPT4all is a community-driven project trained on a massive curated collection of written texts of assistant interactions, including code, stories, depictions, and multi-turn dialogue. The team has provided datasets, model weights, data curation processes, and training code to promote the open-source model. There is also a release of a quantized 4-bit version of the model that is able to run on your laptop as the memory and computation power required is less. A Python client is also available that you can use to interact with the model.

Tags: full Stars: ⭐⭐⭐⭐⭐

oobabooga/text-generation-webui

A gradio web UI for running Large Language Models like GPT-J 6B, OPT, GALACTICA, LLaMA, and Pygmalion.

Tags: Full

KoboldAI/KoboldAI-Client

This is a browser-based front-end for AI-assisted writing with multiple local & remote AI models. It offers the standard array of tools, including Memory, Author’s Note, World Info, Save & Load, adjustable AI settings, formatting options, and the ability to import existing AI Dungeon adventures. You can also turn on Adventure mode and play the game like AI Dungeon Unleashed.

Tags: Full

young-geng/EasyLM

Large language models (LLMs) made easy, EasyLM is a one stop solution for pre-training, finetuning, evaluating and serving LLMs in JAX/Flax. EasyLM can scale up LLM training to hundreds of TPU/GPU accelerators by leveraging JAX's pjit functionality.

Lightning-AI/lit-llama

Implementation of the LLaMA language model based on nanoGPT.

BlinkDL/ChatRWKV

ChatRWKV is an open-source chatbot powered by RWKV, an RNN with Transformer-level LLM performance language model. Model results are comparable with those of ChatGPT. The model uses RNNs. Fine-tuning of the model was done using Stanford Alpaca and other datasets.

Tags: Full

bigscience-workshop/xmtf

This repository provides an overview of all components used for the creation of BLOOMZ & mT0 and xP3 introduced in the paper Crosslingual Generalization through Multitask Finetuning.

Related links:

Tags: Standard

carperai/trlx

A repo for distributed training of language models with Reinforcement Learning via Human Feedback (RLHF), supporting online RL up to 20b params and offline RL to larger models. Basically what you would use to finetune GPT into ChatGPT.

Tags: Bare

LianjiaTech/BELLE

The goal of this project is to promote the development of the open-source community for Chinese language large-scale conversational models. This project optimizes Chinese performance in addition to original Stanford Alpaca. The model finetuning uses only data generated via ChatGPT (without other data). This repo contains: 175 chinese seed tasks used for generating the data, code for generating the data, 0.5M generated data used for fine-tuning the model, model finetuned from BLOOMZ-7B1-mt on data generated by this project.

Related links:

Tags: Standard

ethanyanjiali/minChatGPT

A minimum example of aligning language models with RLHF similar to ChatGPT

Related links:

Tags: Standard

cerebras/Cerebras-GPT

7 open source GPT-3 style models with parameter ranges from 111 million to 13 billion, trained using the Chinchilla formula. Model weights have been released under a permissive license (Apache 2.0 license in particular).

Related links:

Tags: Standard

TavernAI/TavernAI

Atmospheric adventure chat for AI language model Pygmalion by default and other models such as KoboldAI, ChatGPT, GPT-4

Tags: Full

Other LLaMA-derived projects:

  • project-baize/baize-chatbot Baize is an open-source chat model trained with LoRA. It uses 100k dialogs generated by letting ChatGPT chat with itself. We also use Alpaca's data to improve its performance. We have released 7B, 13B and 30B models. Please refer to the paper for more details.
  • pointnetwork/point-alpaca Released weights recreated from Stanford Alpaca, an experiment in fine-tuning LLaMA on a synthetic instruction dataset.
  • tloen/alpaca-lora Code for rproducing the Stanford Alpaca results using low-rank adaptation (LoRA).
  • ggerganov/llama.cpp Ports for inferencing LLaMA in C/C++ running on CPUs, supports alpaca, gpt4all, etc.
  • setzer22/llama-rs Rust port of the llama.cpp project.
  • juncongmoo/chatllama Open source implementation for LLaMA-based ChatGPT runnable in a single GPU.
  • Lightning-AI/lit-llama Implementation of the LLaMA language model based on nanoGPT.
  • nomic-ai/gpt4all Demo, data and code to train an assistant-style large language model with ~800k GPT-3.5-Turbo Generations based on LLaMA.
  • hpcaitech/ColossalAI#ColossalChat An open-source solution for cloning ChatGPT with a complete RLHF pipeline.
  • lm-sys/FastChat An open platform for training, serving, and evaluating large language model based chatbots.
  • nsarrazin/serge A web interface for chatting with Alpaca through llama.cpp. Fully dockerized, with an easy to use API.

About

Open efforts to implement ChatGPT-like models and beyond.

https://jianzhnie.github.io/machine-learning-wiki/#/deep-rl/

chatgptgpt4instruct-gptrlhfllama

License:Apache License 2.0