DeLighT

Closed weights University of Washington,Allen Institute for AI,Facebook AI Research 99M parameters August 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
University of Washington,Allen Institute for AI,Facebook AI Research
Organisation type
Academia,Research collective,Industry
Country
United States of America, France
Published
3 August 2020
Authors
Sachin Mehta, Marjan Ghazvininejad, Srinivasan Iyer, Luke Zettlemoyer, Hannaneh Hajishirzi

What it does

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

Domain
Language
Task
Language modeling, Translation

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
99M

99M (Table 4b)

Training data
103,000,000 tokens

"100K iterations with a context length of 512 and an effective batch size of 64K tokens." 100000*64000/103000000 = 62.14 epochs

Epochs
62.14
Batch size
64,000

"effective batch size of 64K 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
3.8 × 10¹⁸ FLOP

6 FLOP / parameter / token * 99 * 10^6 parameters * 100000 steps * 64000 tokens per batch = 3.8016e+18 FLOP 31330000000000 FLOP / second / GPU * 8 GPUs * 30 hours [assumed based on smaller models reported training time] * 3600 sec / hour * 0.3 [assumed utilization] = 8.120736e+18 FLOP Operation counting method uses less assumptions

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
30 hours

Table 5 reports training time for 54M translation model (23h) it should be more for the 99M language modeling model.

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
Open source

MIT, training and evaluation for WT103: https://github.com/sacmehta/delight/blob/master/readme_files/lm/wikitext103.md

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

"Comparison with state-of-the-art methods on machine translation corpora. DeLighT delivers similar or better performance than state-of-the-art models with fewer parameters."

Record confidence
Likely
Citations
98
Benchmark data
DeLight

Sources

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

Reference
DeLighT: Deep and Light-weight Transformer
Last updated
28 November 2025

What the numbers mean

Background

DeLighT was published by University of Washington,Allen Institute for AI,Facebook AI Research, in United States of America, in August 2020. academia,Research collective,Industry is the category the publisher falls under.

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

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

How it was trained

Producing it required around 3.8 × 10¹⁸ FLOP of arithmetic, on NVIDIA V100, which is a statement about the training budget rather than about inference.

The training set ran to roughly 103,000,000 tokens.

It is tracked in the underlying dataset for one reason in particular: sOTA improvement.

Answers

DeLighT — common questions

01

How much compute was used to train DeLighT?

Around 3.8 × 10¹⁸ 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.

02

What GPU do I need to run DeLighT?

None. DeLighT 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 DeLighT open source?

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

04

How many parameters does DeLighT have?

DeLighT has 99M parameters. 99M (Table 4b). 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 DeLighT?

DeLighT was published by University of Washington,Allen Institute for AI,Facebook AI Research, based in United States of America, categorised as academia,Research collective,Industry.

06

When was DeLighT released?

DeLighT was published in August 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.

07

What is DeLighT used for?

DeLighT works in Language, and is recorded as handling language modeling, Translation. These are the areas it was designed around; they describe intent rather than a hard boundary.

Source

Original publication

Record last updated 28 November 2025

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.