Discriminator-tuned LSTM

Closed weights Samsung R&D Institute Russia 111.9M parameters November 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
Samsung R&D Institute Russia
Organisation type
Industry
Country
Russia
Published
12 November 2018
Authors
Vadim Popov, Mikhail Kudinov

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
111.9M

The largest experiment is the model trained in section 4.3 ("Large scale experiment"). The dataset here has a vocabulary of 100k words. They use a single-layer LSTM with 500 hidden units, with differentiated softmax for the output layer. d_k for distributed softmax varies from 16-150 depending on the relative frequency of the output word. Input layer: 100k * 500 = 50M parameters LSTM: 4 * (500 + 500) * 500 = 2M parameters Output layer: Here we need to assume the structure of differentiated s…

Training data
tokens

4GB of text * 200M words/GB * (0.75 words/token)^-1 = 1,066,666,667 tokens However, it's unclear whether they train over all of this. They break the data into 20MB chunks and say that each "epoch" corresponds to training on one chunk. Training the discriminator and fine-tuning the language model takes 60 and 30 "epochs" each, i.e. (60 + 30) * 0.02 * 200M * 4/3 = 480M tokens They appear to have pre-trained their own LSTM language model on the same data, but they don't say how many "ep…

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.

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
Likely

Sources

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

Reference
Fine-tuning of Language Models with Discriminator
Last updated
28 November 2025

What the numbers mean

About this model

Discriminator-tuned LSTM was published by Samsung R&D Institute Russia, in Russia, in November 2018. It comes out of 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.

Answers

Discriminator-tuned LSTM — common questions

01

What is Discriminator-tuned LSTM used for?

Discriminator-tuned LSTM 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

What GPU do I need to run Discriminator-tuned LSTM?

None. Discriminator-tuned LSTM 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 Discriminator-tuned LSTM open source?

The licensing for Discriminator-tuned LSTM was never recorded in our source data. We treat unstated licensing as closed, because an unrecorded licence is not one to rely on.

04

How many parameters does Discriminator-tuned LSTM have?

Discriminator-tuned LSTM has 111.9M parameters. The largest experiment is the model trained in section 4.3 ("Large scale experiment"). The dataset here has a vocabulary of 100k words. They use a single-layer LSTM with 500 hidden units, with differentiated softmax for the output layer. d_k for distributed softmax varies from 16-150 depending on the relative frequency of the output word. Input layer: 100k * 500 = 50M parameters LSTM: 4 * (500 + 500) * 500 = 2M parameters Output layer: Here we need to assume the structure of differentiated softmax. Let's (somewhat arbitrarily) assume the top 10k words get 150-dim embeddings, the middle 30k get 50-dim embeddings, and the bottom 60k get 16-dim embeddings. Then we have: (10k * 150) + (30k * 50) + (60k * 16) = 3.96M parameters So in total, we have 50M + 2M + 3.96M = 55.96M In other experiments, the authors use the same architecture for the discriminator as for the language model. So total parameter count would appear to be 2 x 55.96M = 111.92M. 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 Discriminator-tuned LSTM?

Discriminator-tuned LSTM was published by Samsung R&D Institute Russia, based in Russia, categorised as industry.

06

When was Discriminator-tuned LSTM released?

Discriminator-tuned LSTM was published in November 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 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.