Skip to content

Latest commit

 

History

History
116 lines (96 loc) · 16.4 KB

Playground.md

File metadata and controls

116 lines (96 loc) · 16.4 KB
Raw

🔥 Playground

Prompt Engineering aims to carefully curate input prompts that can extract the best possible results from Large language models(LLMs).

🌟 TrustGPT can also serve as a playground for everyone's convenience to learn and practice advanced prompt techniques. You can also commit your issues from TrustGPT to this repo page. Thanks a lot.

We will gradually release the following features:

  1. Prompt example
  2. Question answering over your own document
  3. Autonomous agent
  4. Access to various LLMs

As resources are limited, we suggest using this playground for learning and practicing prompt techniques rather than for work. This will help more people access prompt engineering.

🌀 As a prominent example of LLMs, ChatGPT has received widespread attention and skyrocketed in popularity. Nonetheless, in recent years, a significant number of LLMs have emerged, typically several tens of gigabytes in size and trained on massive amounts of textual data. Therefore, there are several alternatives available that we can use to practice prompt techniques using these models.

🤩 These models in the table below are directly accessible via links, The page contains the usage guide and API interface of the model for the convenience of all developers and researchers to expolre and experience. The Checkpoints can also obtained by corresponding links.

Model Type Lab Playgrounds Params(B) Blog/Paper/Github Checkpoints Announced Time
Falcon Decoder TII 🔗 1,7,40 Blog Falcon-40B-instruct, Falcon-7B-instruct,Falcon-RW-1B,Falcon-RW-7B May-23
GPT-J-6B Decoder EleutherAI 🔗 6 Blog GPT-J-6B, GPT4All-J May-23
DLite Decoder EleutherAI 🔗 0.124-1.5 Blog dlite-v2-1_5b May-23
OpenLLaMA Decoder H2O.AI 🔗 3,7 Github OpenLLaMA-7b-preview-300bt May-23
RedPajama-INCITE Decoder Together 🔗 3-7 Blog RedPajama-INCITE May-23
MPT-7B Decoder mosaic 🔗 7 Blog MPT-7B, MPT-7B-Instruct May-23
h2oGPT Decoder EleutherAI 🔗 12-20 Blog h2oGPT May-23
Dolly Decoder EleutherAI 🔗 3,7,12 Blog/Github dolly-v2-12b Apr-23
Pythia Decoder EleutherAI 🔗 0.07-12 Paper/Github pythia 70M - 12B Apr-23
FastChat-T5 Decoder EleutherAI 🔗 3 Blog fastchat-t5-3b-v1.0 Apr-23
StableLM-Alpha Decoder EleutherAI 🔗 3-65 Github StableLM-Alpha Apr-23
oasst-sft-6-llama-30b Decoder HuggingFace 🔗 30 Github - Apr-23
Cerebras-GPT Decoder HuggingFace 🔗 0.111-13 Paper Cerebras-GPT: A Family of Open, Compute-efficient, Large Language Models Mar-23
OpenAssistant(Pythia family) Decoder LAION AI 🔗 12 Paper/Github OA-Pythia-12B-SFT-8, OA-Pythia-12B-SFT-4, OA-Pythia-12B-SFT-1 Apr-23
GPT-4 Decoder OpenAI 🔗 20 Paper - Mar-23
OpenChatKit Decoder Together 🔗 20 Github - Mar-23
Alpaca Decoder Stanford 🔗 7 Github - Mar-23
ChatGPT Decoder OpenAI 🔗 175 Paper - Nov-22
GPT-JT Decoder Together 🔗 6 Github - Nov-22
Flan-T5 Encoder-Decoder Google Research 🔗 11 Paper/Github Flan-T5 Oct-22
Flan-UL2 Encoder-Decoder Google Research 🔗 20 Paper/Github Flan-UL2 Oct-22
CodeGeeX Decoder Tsinghua 🔗 13 Github CodeGeeX register path Sep-22
GLM-130B Encoder-Decoder Tsinghua & Zhipu 🔗 130 Paper/Github - Aug-22
BLOOM(tr11-176B-ml) Decoder BigScience 🔗 176 Github BLOOM Jul-22
PaLM Decoder Google Research 🔗 540 Paper - Apr-22
GPT-NeoX-20B Decoder EleutherAI 🔗 20 Paper GPT-NEOX-20B Apr-22
CodeT5 Encoder-Decoder Salesforce Research Asia 🔗 small:0.06,base:0.22 Paper - Mar-22
ERNIE3.0 Encoder-Decoder Baidu 🔗 10 Paper - Dec-21
CodeX Decoder OpenAI 🔗 12 Paper - Aug-21
RWKV Decoder OpenAI 🔗 0.1-14 Github RWKV, ChatRWKV Aug-21
GPT-3 Decoder OpenAI 🔗 175 Paper - May-20
T5 Encoder-Decoder Google 🔗 11 Paper T5 Oct-19
RoBERTa Encoder MetaAI 🔗 0.355 Paper roberta-series Jul-19
GPT-2 Decoder OpenAI 🔗 1.5 Paper GPT_2 Series Feb-19
BERT Encoder Google 🔗 0.3 Paper BERT Series Oct-18
GPT-1 Decoder OpenAI 🔗 0.117 Paper GPT_1_seriers Jun-18

🤨 The models in the table below all provide pre-trained weights on which developers can fine-tune (without changing the original backbone architecture), and people can visually see the work of a good team of researchers by using the pre-trained weights of the models directly for a good Demo.

Model Type Lab Github Params(B) Paper/Code Announced Time
LLaMA-65B Decoder MetaAI 🔗 65 Paper/Code Feb-23
OPT-IML Decoder MetaAI 🔗 175 Paper/- Dec-22
ERNIE-Code Encoder-Decoder Baidu 🔗 0.56 Paper/- Dec-22
Galactica Decoder MetaAI 🔗 120 Paper/- Nov-22
mT0 Encoder-Decoder BigScience 🔗 13 Paper/- Nov-22
BLOOMZ Decoder BigScience 🔗 176 Paper/- Nov-22
Atlas Encoder-Decoder MetaAI 🔗 11 Paper/- Aug-22
OPT-175B Decoder MetaAI 🔗 175 Paper/- May-22
RETRO Encoder-Decoder DeepMind 🔗 7.5 Paper/- Dec-21
FLAN Encoder-Decoder Google 🔗 137 Paper/- Sep-21

😣 The following table show that the related models and codes are not open-source till now.

Model Type Lab Report Params(B) Paper/Code Announced Time
Med-PaLM Encoder Google & DeepMind 🔗 540 Paper/- Dec-22
GLaM Encoder Google Inc 🔗 1200 Paper/- Dec-22
RL-CAI Encoder Anthropic 🔗 52 Paper/- Dec-22
Sparrow Decoder DeepMind 🔗 70 Paper/- Sep-22
PaLI Encoder-Decoder Google 🔗 17 Paper/- Sep-22
Gato(Cat) Encoder-Decoder DeepMind 🔗 1 Paper/- May-22
Chinchilla Encoder DeepMind 🔗 70 Paper/- Mar-22
Gopher Encoder DeepMind 🔗 280 Paper/- Dec-21
LaMDA Decoder GoogleAI 🔗 137 Paper/- Jun-21

🎭 The following table show that the LLMs for Code.

Model Checkpoints Paper/Blog Params (B) Announced Time
StarCoder starcoder Blog 15 May-23
StarChat Alpha starchat-alpha Blog 16 May-23
Replit Code replit-code-v1-3b Blog 2.7 May-23
CodeT5+ CodeT5+ Paper 0.22 - 16 May-23
CodeGen2 codegen2 1B-16B Paper 1 - 16 Apr-23
SantaCoder santacoder Paper 1.1 Jan-23

📈 The following table show that the Dataset of LLM area, with instruction-tunning and alignment-tuning.

Dataset Paper/Blog Dataset Samples (K) Announced Time Type
MPT-7B-Instruct Blog dolly_hhrlhf 59 May-23 instruction-tuning
databricks-dolly-15k Blog databricks-dolly-15k 15 Apr-23 instruction-tuning
OpenAssistant Conversations Dataset Blog oasst1 161 Apr-23 alignment-tuning
OIG (Open Instruction Generalist) Blog OIG 44,000 Mar-23 instruction-tuning