LSTM (Hebbian, Cache, MbPA)

Closed weights DeepMind,University College London (UCL) 530.4M parameters March 2018

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
DeepMind,University College London (UCL)
Organisation type
Industry,Academia
Country
United Kingdom of Great Britain and Northern Ireland
Published
27 March 2018
Authors
Jack W Rae, Chris Dyer, Peter Dayan, Timothy P Lillicrap

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
530.4M

Single layer LSTM with hidden dimension of 2048. Vocabulary for Gutenberg is 242,621; input and output embeddings are tied. Embedding layer (tied): 242,621 * 2048 = 496,887,808 LSTM layer: 4 * (2048 + 2048) * 2048 = 33,554,432 Total: 496,887,808 + 33,554,432 = 530,442,240

Training data
175,181,505 tokens

Omniglot: 32k images Wikitext-103: "Over 100 million tokens" Gutenberg: 175,181,505 tokens GigaWord v5: 4B tokens Gigaword is the largest dataset, but the largest training run uses Project Gutenberg.

Epochs
80
Batch size
51,200

Sequence length of 100, total of 512 batches. Batches are split between 8 GPUs.

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
3.3 × 10¹⁹ FLOP

They do training runs on a vision task and three language datasets. The largest dataset by size is GigaWord, but the largest training run is on the Gutenberg dataset, at 15B tokens. I assume the input embedding is done with an embedding lookup for efficiency rather than a dense matrix multiplication, so we only count FLOPs on the de-embedding. Ops counting: 6 * 15B * 530,442,240 = 4.774e19 Hardware: (8 * 1.87e13) * (6 * 24 * 3600) * 0.3 = 2.327e19 Geometric mean: sqrt(4.774e19 * 2.327e19) =…

How it was established
Hardware,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 P100
Chips used
8
Wall-clock time
144 hours

6 days

Power draw
4.2 kW
Compute cost
$591

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

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
47
Benchmark data
LSTM (Hebbian, Cache, MbPA)

Sources

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

Reference
Fast Parametric Learning with Activation Memorization
Last updated
11 February 2026

What the numbers mean

About this model

LSTM (Hebbian, Cache, MbPA) was published by DeepMind,University College London (UCL), in United Kingdom of Great Britain and Northern Ireland, in March 2018. The organisation is categorised as industry,Academia.

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.

How it was trained

Producing it required around 3.3 × 10¹⁹ FLOP of arithmetic, on NVIDIA P100, which is a statement about the training budget rather than about inference.

Around 175,181,505 tokens went into training it.

Answers

LSTM (Hebbian, Cache, MbPA) — common questions

01

What is LSTM (Hebbian, Cache, MbPA) used for?

LSTM (Hebbian, Cache, MbPA) 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.

02

How much compute was used to train LSTM (Hebbian, Cache, MbPA)?

Around 3.3 × 10¹⁹ FLOP, on NVIDIA P100. 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.

03

What GPU do I need to run LSTM (Hebbian, Cache, MbPA)?

None. LSTM (Hebbian, Cache, MbPA) 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.

04

Is LSTM (Hebbian, Cache, MbPA) open source?

No. LSTM (Hebbian, Cache, MbPA) has not had its weights published, so it exists only as a service controlled by its owner.

05

How many parameters does LSTM (Hebbian, Cache, MbPA) have?

LSTM (Hebbian, Cache, MbPA) has 530.4M parameters. Single layer LSTM with hidden dimension of 2048. Vocabulary for Gutenberg is 242,621; input and output embeddings are tied. Embedding layer (tied): 242,621 * 2048 = 496,887,808 LSTM layer: 4 * (2048 + 2048) * 2048 = 33,554,432 Total: 496,887,808 + 33,554,432 = 530,442,240. 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.

06

Who created LSTM (Hebbian, Cache, MbPA)?

LSTM (Hebbian, Cache, MbPA) was published by DeepMind,University College London (UCL), based in United Kingdom of Great Britain and Northern Ireland, categorised as industry,Academia.

07

When was LSTM (Hebbian, Cache, MbPA) released?

LSTM (Hebbian, Cache, MbPA) was published in March 2018. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

Source

Original publication

Record last updated 11 February 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.