KEPLER

Closed weights Tsinghua University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),HEC,CIFAR AI Research,Princeton University,University of Montreal / Université de Montréal 125M parameters November 2020

No estimate

No hardware requirements for this model

The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
Tsinghua University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),HEC,CIFAR AI Research,Princeton University,University of Montreal / Université de Montréal
Organisation type
Academia,Academia,Academia,Research collective,Academia,Academia
Country
China, Canada, France, United States of America
Published
23 November 2020
Authors
Xiaozhi Wang, Tianyu Gao, Zhaocheng Zhu, Zhiyuan Liu, Juanzi Li, and Jian Tang.

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Language
Task
Relation extraction
Approach
Self-supervised learning
Base model
RoBERTa Base
Numerical format
FP16

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
125M
Training data
3,533,640,000 tokens

For BookCorpus + English Wikipedia: 800M + 2500M For Wikidata5M: 20614279 See table 1. Contains "entities", "relations", and "triplets"

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute
1.7 × 10²¹ FLOP

From author communication "About 128 GPU-days using Nvidia V100 (16GB). " precision: float16 V100 GPU for float16: 28000000000000 (2.8E+13) 0.4 * 28TFLOP/s * 128 GPU-days * 24h/day * 3600s/h = 1.24E+20 "and use the released roberta.base parameters for initialization, which is a common practice to save pre-training time" Roberta base FLOP: 1.536e+21 Total:1.660000e+21

How it was established
Hardware

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights
Closed — provider access only
Model access
Unreleased
Training code
Open source

MIT License, includes train code https://github.com/THU-KEG/KEPLER

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Why it is tracked
SOTA improvement

"Experimental results show that KEPLER achieves state-of-the-art performances on various NLP tasks"

Record confidence
Confident
Citations
819

Sources

Where this record came from and when it was last checked.

Reference
KEPLER: A Unified Model for Knowledge Embedding and Pre- trained Language Representation.
Last updated
25 May 2026

What the numbers mean

About this model

KEPLER was published by Tsinghua University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),HEC,CIFAR AI Research,Princeton University,University of Montreal / Université de Montréal, in China, in November 2020. academia,Academia,Academia,Research collective,Academia,Academia is the category the publisher falls under.

It works in Language, and is recorded as doing relation extraction.

It builds on RoBERTa Base, which is why it shares that model's general shape and size.

Its weights were never published, so it can only be reached through its provider. No graphics card changes that.

What went into building it

Training it took roughly 1.7 × 10²¹ FLOP of computation — a measure of what producing the model cost, not of how fast it answers.

Around 3,533,640,000 tokens went into training it.

Its inclusion criterion is sOTA improvement.

Answers

KEPLER — common questions

01

When was KEPLER released?

KEPLER was published in November 2020. 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 KEPLER used for?

KEPLER works in Language, and is recorded as handling relation extraction. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

03

How much compute was used to train KEPLER?

Around 1.7 × 10²¹ FLOP. 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 KEPLER?

None. KEPLER 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 KEPLER open source?

No. KEPLER has not had its weights published, so it exists only as a service controlled by its owner.

06

How many parameters does KEPLER have?

KEPLER has 125M parameters. 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 KEPLER?

KEPLER was published by Tsinghua University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),HEC,CIFAR AI Research,Princeton University,University of Montreal / Université de Montréal, based in China, categorised as academia,Academia,Academia,Research collective,Academia,Academia.

Source

Original publication

Record last updated 25 May 2026

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.