News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM

Dense decoder transformers

The GPT design: one stack of transformer layers predicting the next token. 308 models on EveryLLM so far.

The dense decoder transformer is the design behind GPT and most chatbots. It is a single stack of identical transformer layers that reads the tokens so far and predicts the next one, with a mask that stops each position from looking ahead. Dense means every parameter is used for every token, unlike in a mixture of experts. GPT-3, PaLM, Llama 3, Chinchilla, Gemma, Phi and OLMo are dense decoders.

Inside each layer

Each layer has two parts. Self-attention lets every token weigh every earlier token and gather what it needs; a feed-forward network then transforms each token's vector on its own. The transformer was introduced by Google researchers in the June 2017 paper Attention Is All You Need, for translation, and OpenAI's GPT-1 (2018) kept only its decoder half. Llama 3.1 405B, released in July 2024, stacks 126 such layers.

Why it won

A decoder trains on any text without labels, because the next token is always the answer to predict, and it scales smoothly: scaling laws published by OpenAI in 2020 and DeepMind in 2022 showed the error falling predictably with more parameters, data and compute. Its weak point is cost. Attention grows with the square of the context length, and every parameter must be read for every token, which is why mixtures of experts became popular.

Famous dense decoder models

Still to be written: OLMo (language model).

Further reading

XLNet WorldXLNet Permutation-based autoregressive transformer model created by Google AIResearch · Open weights · 2019 GPT-J WorldGPT-J A 6-billion-parameter open English text generator from EleutherAI, released in 2021.General · Open weights · 2021 1.58-bit large language model World1.58-bit large language model A model whose weights hold only three values: -1, 0 and +1.Research · Open weights · 2024 GPT-1 WorldGPT-1 OpenAI's first GPT model, a 117M-parameter decoder-only transformer from 2018.Research · Open weights · 2018 GPT-2 🇺🇸 United StatesGPT-2 OpenAI's 2019 text model that predicts the next word in a sequenceGeneral · Open weights · 2019 Talkie 🇺🇸 United StatesTalkie A 13B small language model trained only on pre-1931 public domain English textSmall · Open weights · 2026 INTELLECT-1 WorldINTELLECT-1 A 10 billion parameter model trained by a global community.General · Open weights · 2024 OpenELM 🇺🇸 United StatesOpenELM Apple's small open language models with a full training and inference releaseSmall · Open weights · 2024 DialoGPT 🇺🇸 United StatesDialoGPT A GPT-style model trained on Reddit threads to write chat repliesGeneral · Open weights · 2019 BioGPT 🇺🇸 United StatesBioGPT A generative Transformer trained on biomedical writing for text and mining tasksDomain · Open weights · 2022 Ganga-1B 🇮🇳 IndiaGanga-1B Hindi-focused autoregressive language model from IIT GandhinagarRegional · Open weights Cerebras-GPT 🇺🇸 United StatesCerebras-GPT Open compute-optimal language models from Cerebras SystemsGeneral · Open weights · 2023 GPT-NeoX-20B 🇺🇸 United StatesGPT-NeoX-20B Open 20 billion parameter autoregressive language model by EleutherAIGeneral · Open weights · 2022 Tülu 3 🇺🇸 United StatesTülu 3 Open post-trained model family built on Llama 3.1General · Open weights · 2024 RuGPT-3 🇷🇺 RussiaRuGPT-3 Russian language model based on GPT-2 architectureRegional · Open weights imgCogVLM 🇨🇳 ChinaCogVLM Open-source visual language model with deep feature fusionMultimodal · Open weights · 2023 Phi-2 🇺🇸 United StatesPhi-2 A 2.7 billion parameter Microsoft research model for QA, chat and codeResearch · Open weights Pythia 🇺🇸 United StatesPythia A suite of 16 models for interpretability research.Research · Open weights · 2023 ULMFiT WorldULMFiT Transfer learning method for NLP text classification tasksResearch · Open weights · 2018

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.