GPT-4 (Jun 2023)

Uzavřené váhy OpenAI 1.8T Parametry June 2023

žádný odhad

Žádné hardwarové požadavky pro tento model

Váhy pro tento model nebyly zveřejněny, takže ho nelze stáhnout ani spustit na vašem vlastním hardwaru v jakékoli velikosti. Je dostupný pouze přes jeho poskytovatele a žádná grafická karta to nezmění..

Na záznamu

Plná specifikace

Všechno je zaznamenáno pro tento model. Většina z toho popisuje, jak byl trénován, spíše než jak běží — užitečný kontext pro posouzení, kolik práce do toho bylo vloženo, a jak se to srovnává s modely vytvořenými v jiném měřítku..

Origin

Kdo vytvořil tento model, kde a kdy byl publikován.

Organizace
OpenAI
Typ organizace
Industry
Stát
United States of America
Published
13 June 2023
Autoři
OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, Irwan Bello, Jake Berdine, Gabriel Bernadett-Shapiro, Christopher Berner, Lenny Bogdonoff, Oleg Boiko, Madelaine Bo…

Co to dělá

Problémové oblasti, pro které byl model navržen. Model může mít několik z každé.

Domain
Multimodal, Language, Vision
Úloha
Language modeling, Language modeling/generation, Question answering, Visual question answering
Přístup
Self-supervised learning
Base model
GPT-4 (Mar 2023)

Velikost

Jak velký je model a na kolik dat byl trénován. Parametry jsou číslo, které rozhoduje, zda se vejde na danou grafickou kartu.

Parametry
1.8T

Rumored to be 1.8T parameter MoE with 280B activated on the forward pass, per https://www.semianalysis.com/p/gpt-4-architecture-infrastructure. Other sources estimate 1.76T with 220B per forward pass https://web.archive.org/web/20230712123915/https://the-decoder.com/gpt-4-architecture-datasets-costs-and-more-leaked/

Tréninková data
5,416,666,666,667 tokens

Speculative. Reported secondhand by online sources such as Semianalysis, but not verified by OpenAI. If total number of tokens seen was 13T, text was repeated for 2 epochs, and text was the majority of tokens, then dataset size roughly is 13T*0.75/2 = 4.9T words. Note this examines only the text dataset, since GPT-4 was first and foremost a language model. However, the vision component had its own vision dataset, which we believe accounted for a much smaller part of the compute budget.

Epochs
2
Batch size
60,000,000

MoE so i hesitate but 7.5 mill? https://www.reddit.com/r/mlscaling/comments/14wcy7m/gpt4s_details_are_leaked/#:~:text=There%20is%20millions%20of%20rows,get%20the%20real%20batch%20size.

Výpočet trénování

Aritmetika prováděná k trénování modelu, měřená ve floating-point operacích. Je to měřítko toho, kolik stálo trénování, nikoli jak rychle hotový model odpovídá.

Výpočet trénování
2.1 × 10²⁵ FLOP

90% CI: 8.2E+24 to 4.4E+25 NOTE: this is a rough estimate based on public information, much less information than most other systems in the database. Calculation and confidence intervals here: https://colab.research.google.com/drive/1O99z9b1I5O66bT78r9ScslE_nOj5irN9?usp=sharing

How it was established
Hardware

Tréninkový běh

Co to fyzicky stálo na trénink: které čipy, kolik, jak dlouho a kolik to bralo ze zásuvky.

Tréninkový hardware
NVIDIA A100 SXM4 40 GB
Chips used
25,000
Chip-hours
57,000,000
Wall-clock time
2,280 hours (95 days)

(Speculative) SemiAnalysis conjectures that GPT-4 training took 90-100 days with utilization of 32-36%.

Hardware utilisation
MFU 34.0%

CANNOT VERIFY, LIKELY HFU, PAYWALLED. (Speculative) SemiAnalysis conjectures that GPT-4 had utilization of 32-36%: https://www.semianalysis.com/p/gpt-4-architecture-infrastructure

Zatížení energie
19.9 MW

Dostupnost

Zda můžete získat model a spustit ho na vlastním hardwaru, což rozhoduje o tom, zda se zde uvedené hodnoty grafické karty vztahují.

Váhy
Closed — provider access only
Přístup k modelu
API access
Training code
Unreleased

Jak je to klasifikováno

Štítky, které se aplikují na zdrojový dataset při sledování významných modelů a jakou má jistotu v tomto záznamu.

Frontier model
Yes
Foundation model
Yes
Likely above 10²³ FLOP
Yes
Proč je to sledováno
Highly cited,SOTA improvement,Training cost

See the paper, p.1: "On a suite of traditional NLP benchmarks, GPT-4 outperforms both previous large language models and most state-of-the-art systems (which often have benchmark-specific training or hand-engineering)." "On the MMLU benchmark [35, 36], an English-language suite of multiple-choice questions covering 57 subjects, GPT-4 not only outperforms existing models by a considerable margin in English, but also demonstrates strong performance in other languages"

Model/modely nelze spustit. Tightly = téměř nelze spustit, comfortable = snadno spustitelné s rezervou.
Likely
model/nebudou vyhovovat běhu AI modelu
24,490

Model/modely nevyhovuje / nebudou vyhovovat. Těsný = barely runnable, pohodlný = běží snadno s rezervou.

Kde tento záznam pochází a kdy byl naposledy zkontrolován.

Odkaz
GPT-4 Technical Report
Poslední aktualizace
24 June 2026

Co znamenají čísla

Co je tento model

GPT-4 (Jun 2023) was published by OpenAI, in United States of America, in June 2023. It comes out of industry.

It works in Multimodal, Language, Vision, and is recorded as doing language modeling, Language modeling/generation, Question answering, Visual question answering.

It builds on GPT-4 (Mar 2023), which is why it shares that model's general shape and size.

Jeho váhy nikdy nebyly zveřejněny, takže je možné k němu přistupovat jedině prostřednictvím jeho poskytovatele. Žádná grafická karta to nezmění.

How it was trained

The training run consumed about 2.1 × 10²⁵ FLOP, on NVIDIA A100 SXM4 40 GB. That figure describes the cost of creating it and has no bearing on how quickly it generates text.

Around 5,416,666,666,667 tokens went into training it.

It is tracked in the underlying dataset for one reason in particular: highly cited,SOTA improvement,Training cost.

GPT-4 (Jun 2023) — Běžné otázky

01

When was GPT-4 (Jun 2023) released?

GPT-4 (Jun 2023) was published in June 2023. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

02

What is GPT-4 (Jun 2023) used for?

GPT-4 (Jun 2023) works in Multimodal, Language, Vision, and is recorded as handling language modeling, Language modeling/generation, Question answering, Visual question answering. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

03

How much compute was used to train GPT-4 (Jun 2023)?

Around 2.1 × 10²⁵ FLOP, on NVIDIA A100 SXM4 40 GB. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

04

What GPU do I need to run GPT-4 (Jun 2023)?

None. GPT-4 (Jun 2023) is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.

05

Is GPT-4 (Jun 2023) open source?

No. GPT-4 (Jun 2023) has not had its weights published, so it exists only as a service controlled by its owner.

06

How many parameters does GPT-4 (Jun 2023) have?

GPT-4 (Jun 2023) has 1.8T parameters. Rumored to be 1.8T parameter MoE with 280B activated on the forward pass, per https://www.semianalysis.com/p/gpt-4-architecture-infrastructure. Other sources estimate 1.76T with 220B per forward pass https://web.archive.org/web/20230712123915/https://the-decoder.com/gpt-4-architecture-datasets-costs-and-more-leaked/. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

07

Who created GPT-4 (Jun 2023)?

GPT-4 (Jun 2023) was published by OpenAI, based in United States of America, categorised as industry.