T2R 75% + Pretrain (WT-103)

Closed weights University of Washington,Microsoft,DeepMind 668.9M parameters March 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 Washington,Microsoft,DeepMind
Organisation type
Academia,Industry,Industry
Country
United States of America, United Kingdom of Great Britain and Northern Ireland
Published
24 March 2021
Authors
Jungo Kasai, Hao Peng, Yizhe Zhang, Dani Yogatama, Gabriel Ilharco, Nikolaos Pappas, Yi Mao, Weizhu Chen, Noah A. Smith

What it does

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

Domain
Language
Task
Language modeling

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
668.9M

Vocabulary size unclear, ~260k if they directly use WT 103 Assumed tied embeddings Remaining parameters: 32 layers 8 attention heads (H) 128 head dimensions (N, D) 1024 model (embedding) dimensions (W, M) 4096 feedforward dimensions "We partition the training data into non-overlapping blocks of 512 contiguous tokens" MLP layer: 2*1024*4096=8388608 Att layer: H×(W×(2×D + N) + N×M) 8*(1024*(2*128+128)+128*1024)=4194304 Total non embedding: 32*(8388608+4194304)=402653184 Embedding: 260000*1024…

Training data
tokens
Epochs
34.47

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
1.4 × 10¹⁹ FLOP

Table 4, "We found that we could speed up training by reducing the warm-up steps, total update steps, maximum and minimum rates, and batch size to 8K steps, 142K steps, 5 ⋅ 10−6, 0.5, and 25K tokens " 142K updates, 25k token batches Training compute: 6*668893184*142000*25000=1.4247425e+19 GPU hour estimate: 95*60*60*125000000000000*0.3=1.2825e+19 Geometric mean: 13517515512289972224 Epochs: 142000*25000/103000000=34.4660194175

How it was established
Operation counting,Hardware

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Training hardware
NVIDIA V100
Chips used
8
Wall-clock time
12 hours

95h GPU time / 8 GPUs ~11.8h wall clock time

Power draw
4.9 kW

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
Unreleased

There's a repo but it's kind of inscrutable with no docs about T2R, not clear if the training code for this paper is in it: https://github.com/jungokasai/T2R/tree/master

How it is classified

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

Record confidence
Confident
Citations
94
Benchmark data
T2R 75% + Pretrain

Sources

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

Reference
Finetuning Pretrained Transformers into RNNs
Last updated
25 May 2026

What the numbers mean

About this model

T2R 75% + Pretrain (WT-103) was published by University of Washington,Microsoft,DeepMind, in United States of America, in March 2021. The organisation is categorised as academia,Industry,Industry.

It works in Language, and is recorded as doing language modeling.

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

What went into building it

The training run consumed about 1.4 × 10¹⁹ FLOP, on NVIDIA V100. That figure describes the cost of creating it and has no bearing on how quickly it generates text.

Answers

T2R 75% + Pretrain (WT-103) — common questions

01

When was T2R 75% + Pretrain (WT-103) released?

T2R 75% + Pretrain (WT-103) was published in March 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.

02

What is T2R 75% + Pretrain (WT-103) used for?

T2R 75% + Pretrain (WT-103) works in Language, and is recorded as handling language modeling. 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 T2R 75% + Pretrain (WT-103)?

Around 1.4 × 10¹⁹ FLOP, on NVIDIA V100. 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 T2R 75% + Pretrain (WT-103)?

None. T2R 75% + Pretrain (WT-103) 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 T2R 75% + Pretrain (WT-103) open source?

No. T2R 75% + Pretrain (WT-103) has not had its weights published, so it exists only as a service controlled by its owner.

06

How many parameters does T2R 75% + Pretrain (WT-103) have?

T2R 75% + Pretrain (WT-103) has 668.9M parameters. Vocabulary size unclear, ~260k if they directly use WT 103 Assumed tied embeddings Remaining parameters: 32 layers 8 attention heads (H) 128 head dimensions (N, D) 1024 model (embedding) dimensions (W, M) 4096 feedforward dimensions "We partition the training data into non-overlapping blocks of 512 contiguous tokens" MLP layer: 2*1024*4096=8388608 Att layer: H×(W×(2×D + N) + N×M) 8*(1024*(2*128+128)+128*1024)=4194304 Total non embedding: 32*(8388608+4194304)=402653184 Embedding: 260000*1024=266240000 Total: 402653184+266240000=668893184. 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 T2R 75% + Pretrain (WT-103)?

T2R 75% + Pretrain (WT-103) was published by University of Washington,Microsoft,DeepMind, based in United States of America, categorised as academia,Industry,Industry.

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.