DiffQ Transformer (16L)

Closed weights Meta AI 247M parameters April 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
Meta AI
Organisation type
Industry
Country
United States of America
Published
20 April 2021
Authors
Alexandre Défossez, Yossi Adi, Gabriel Synnaeve

What it does

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

Domain
Language
Task
Language modeling
Approach
Self-supervised learning

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
247M

They base their architecture off of "Transformer (Adaptive Input Embeddings) WT103", a 16 layer transformer with 16 heads per MHA layer, d_model=1024 and d_ff=4096.

Training data
103,000,000 tokens

WikiText-103, which has 103M training 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 × 10¹⁹ FLOP

The authors say they train their base model following Baevski & Auli (2019). That paper trained for 286k steps in batches of 65,536 tokens. This suggests training the base model took 6 * 247M * 286k * 65536 = 2.778E19 FLOPs. However, DiffQ introduces some additional computational overhead during training: "Using DiffQ usually increase the training time by some amount. On the language modeling task, the time per batch went from 115ms to 125ms." This suggests 125/115 - 1 = 8.7% overhead, relati…

How it was established
Operation counting

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

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 (non-commercial)

training code (non-commercial): https://github.com/facebookresearch/diffq/blob/main/examples/FAIRSEQ_README.md

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
63
Benchmark data
DiffQ Transformer (16L)

Sources

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

Reference
Differentiable Model Compression via Pseudo Quantization Noise
Last updated
25 May 2026

What the numbers mean

Background

DiffQ Transformer (16L) was published by Meta AI, in United States of America, in April 2021. industry is the category the publisher falls under.

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

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

What went into building it

The training run consumed about 3 × 10¹⁹ FLOP, on NVIDIA V100. That figure describes the cost of creating it and has no bearing on how quickly it generates text.

Around 103,000,000 tokens went into training it.

Answers

DiffQ Transformer (16L) — common questions

01

How many parameters does DiffQ Transformer (16L) have?

DiffQ Transformer (16L) has 247M parameters. They base their architecture off of "Transformer (Adaptive Input Embeddings) WT103", a 16 layer transformer with 16 heads per MHA layer, d_model=1024 and d_ff=4096. 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.

02

Who created DiffQ Transformer (16L)?

DiffQ Transformer (16L) was published by Meta AI, based in United States of America, categorised as industry.

03

When was DiffQ Transformer (16L) released?

DiffQ Transformer (16L) was published in April 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.

04

What is DiffQ Transformer (16L) used for?

DiffQ Transformer (16L) works in Language, and is recorded as handling language modeling. These are the areas it was designed around; they describe intent rather than a hard boundary.

05

How much compute was used to train DiffQ Transformer (16L)?

Around 3 × 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.

06

What GPU do I need to run DiffQ Transformer (16L)?

None. DiffQ Transformer (16L) 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.

07

Is DiffQ Transformer (16L) open source?

No. DiffQ Transformer (16L) has not had its weights published, so it exists only as a service controlled by its owner.

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.