TOME

Closed weights University of Southern California,Google 220M parameters October 2021

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
University of Southern California,Google
Organisation type
Academia,Industry
Country
United States of America
Published
12 October 2021
Authors
Michiel de Jong, Yury Zemlyanskiy, Nicholas FitzGerald, Fei Sha, William Cohen

What it does

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

Domain
Language
Task
Question answering
Numerical format
FP32

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
220M

220M from Table 1 entry for TOME

Training data
tokens

Per Wikipedia, the site had 3.755B words as of the most recent checkpoint prior to the paper's publication: https://en.wikipedia.org/wiki/Wikipedia:Size_of_Wikipedia#Yearly_statistics 3.755B * 4/3 = 5B tokens

Epochs
157.29
Batch size
524,288

"All models are pre-trained on [...] batch size of 4096. Each passage in the batch has length T = 128, excluding entity tokens."

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 × 10²¹ FLOP

We have information about hardware "All models are pre-trained on 128 TPUs using AdamW optimizer (Loshchilov& Hutter, 2019) with learning rate 1e-4 and batch size of 4096.", but no information about training time and exact model of TPUs. There is information about training 1.5M steps with batch size 4096 * 128 citation from appendix A: "The Mention Encoder and BATCH-TOME are pre-trained for 1 million steps with 50k warmup steps, and TOME is trained for 500k additional steps with 25k warmup step…

How it was established
Operation counting

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

Apache 2.0 (repo license) https://github.com/google-research/language/tree/master/language/mentionmemory

How it is classified

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

Record confidence
Likely
Citations
55

Sources

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

Reference
Mention Memory: incorporating textual knowledge into Transformers through entity mention attention
Last updated
25 May 2026

What the numbers mean

Background

TOME was published by University of Southern California,Google, in United States of America, in October 2021. The organisation is categorised as academia,Industry.

It works in Language, and is recorded as doing question answering.

This is a closed model: the trained values stayed with whoever produced them, and there is no local version to run.

How it was trained

Producing it required around 1 × 10²¹ FLOP of arithmetic, which is a statement about the training budget rather than about inference.

Answers

TOME — common questions

01

How much compute was used to train TOME?

Around 1 × 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.

02

What GPU do I need to run TOME?

None. TOME 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.

03

Is TOME open source?

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

04

How many parameters does TOME have?

TOME has 220M parameters. 220M from Table 1 entry for TOME. 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.

05

Who created TOME?

TOME was published by University of Southern California,Google, based in United States of America, categorised as academia,Industry.

06

When was TOME released?

TOME was published in October 2021. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

07

What is TOME used for?

TOME works in Language, and is recorded as handling question answering. These are the areas it was designed around; they describe intent rather than a hard boundary.

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.