EleutherAI/gpt-neox

• Bronze V

An implementation of model parallel GPT-2 and GPT-3-style models using the mesh-tensorflow library.

• 1,372+2 Elo, change today
• 8.2k stars
• 955 forks
• 4y ago
• archived

git clone https://github.com/EleutherAI/gpt-neo.git

• gpt
• gpt-2
• gpt-3
• language-model

• Silver III

RWKV (pronounced RwaKuv) is an RNN with great LLM performance, which can also be directly trained like a GPT transformer (parallelizable). We are at RWKV-7 "Goose". So it's combining the best of RNN and transformer - great performance, linear time, constant space (no kv-cache), fast training, infinite ctx_len, and free sentence embedding.

• 1,475−6 Elo, change today
• 14.7k stars
• 1k forks
• 11d ago

git clone https://github.com/BlinkDL/RWKV-LM.git

• attention-mechanism
• chatgpt
• deep-learning
• gpt

• Gold III

💥 Fast State-of-the-Art Tokenizers optimized for Research and Production

• 1,578+3 Elo, change today
• 11.2k stars
• 1.2k forks
• 8h ago

git clone https://github.com/huggingface/tokenizers.git

• bert
• gpt
• language-model
• natural-language-processing

• Silver III

Adding guardrails to large language models.

• 1,470+2 Elo, change today
• 7.6k stars
• 742 forks
• 4d ago

git clone https://github.com/guardrails-ai/guardrails.git

• ai
• foundation-model
• gpt-3
• llm

• Gold IV

Code for loralib, an implementation of "LoRA: Low-Rank Adaptation of Large Language Models"

• 1,550−7 Elo, change today
• 13.9k stars
• 936 forks
• 1y ago

git clone https://github.com/microsoft/LoRA.git

• adaptation
• deberta
• deep-learning
• gpt-2

• Gold I

Code and documentation to train Stanford's Alpaca models, and generate the data.

• 1,610−6 Elo, change today
• 30.2k stars
• 3.9k forks
• 2y ago

git clone https://github.com/tatsu-lab/stanford_alpaca.git

• deep-learning
• instruction-following
• language-model