LSTM+NeuralCache

Closed weights KU Leuven,ESAT - PSI,Apple 2.1M parameters September 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
KU Leuven,ESAT - PSI,Apple
Organisation type
Academia,Academia,Industry
Country
Belgium, United States of America
Published
24 September 2018
Authors
Lyan Verwimp, Joris Pelemans, Hugo Van hamme, Patrick Wambacq

What it does

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

Domain
Language
Task
Language modeling
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
2.1M

Given: Hidden size (H) = 512 Number of hidden layers (L) = 1 Input size (I) is not mentioned, so let's denote it as I The total number of parameters P in an LSTM can be calculated as follows: P = 4 * ((I * H) + (H * H) + H) This is for one layer of LSTM cells. Since the LSTM model described has only one layer, we don't need to multiply by the number of layers. To calculate the exact number of parameters, we would need to know the input size I. However, if you are looking for the number of pa…

Training data
2,000,000 tokens

" we stop training anyway after 39 epochs"

Epochs
39

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
9.8 × 10¹⁴ FLOP

6 FLOP / parameter / token * 2100000 parameters * 2000000 tokens * 39 epochs = 9.828e+14 FLOP

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
Unreleased

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

"We obtain a 29.9%/32.1% (validation/test set) relative improvement in perplexity with respect to a baseline LSTM LM on the WikiText-2 dataset, outperforming previous work on neural cache LMs" ... "we observe that neural cache models consistently outperform regular cache models on this dataset."

Record confidence
Likely
Citations
3
Benchmark data
LSTM+NeuralCache

Sources

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

Reference
Information-Weighted Neural Cache Language Models for ASR
Last updated
28 November 2025

What the numbers mean

Background

LSTM+NeuralCache was published by KU Leuven,ESAT - PSI,Apple, in the country recorded as Belgium, during September 2018. The category the publisher falls under is academia,Academia,Industry.

It works in the domain of Language, and is recorded as performing the task of language modeling.

Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.

Training and provenance

Producing it required arithmetic totalling around 9.8 × 10¹⁴ FLOP. That figure measures what producing the model cost, and has no bearing on how fast it answers.

The training set ran to roughly 2,000,000 tokens of text.

The reason it appears in this catalogue at all: sOTA improvement.

Answers

LSTM+NeuralCache — common questions

01

LSTM+NeuralCache— how much compute was used to train it?

Training consumed around 9.8 × 10¹⁴ 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

LSTM+NeuralCache— what GPU do I need to run it?

None. This 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

LSTM+NeuralCache— is it open source?

No. Its weights have not been published, so it exists only as a service controlled by its owner.

04

LSTM+NeuralCache— how many parameters does it have?

It has a parameter count of 2.1M. Given: Hidden size (H) = 512 Number of hidden layers (L) = 1 Input size (I) is not mentioned, so let's denote it as I The total number of parameters P in an LSTM can be calculated as follows: P = 4 * ((I * H) + (H * H) + H) This is for one layer of LSTM cells. Since the LSTM model described has only one layer, we don't need to multiply by the number of layers. To calculate the exact number of parameters, we would need to know the input size I. However, if you are looking for the number of parameters just within a single LSTM cell (assuming I is equal to H), then you can substitute I with H in the above formula: P = 4 * ((H * H) + (H * H) + H) = 4 * (2 * (H^2) + H) For H = 512, this becomes: P = 4 * (2 * (512^2) + 512) = 4 * (2 * 262144 + 512) = 4 * (524288 + 512) = 4 * 524800 ≈ 2,099,200. 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

LSTM+NeuralCache— who created it?

It was published by KU Leuven,ESAT - PSI,Apple, based in Belgium, an organisation categorised as academia,Academia,Industry.

06

LSTM+NeuralCache— when was it released?

It was published in September 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.

07

LSTM+NeuralCache— what is it used for?

It works in the domain of Language, and is recorded as handling the task of 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.

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.