WeNet (Penn Treebank)

Closed weights Amazon 23M parameters April 2019

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
Amazon
Organisation type
Industry
Country
United States of America
Published
8 April 2019
Authors
Zhiheng Huang, Bing Xiang

What it does

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

Domain
Language
Task
Neural Architecture Search - NAS, Language modeling
Numerical format
FP16

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
23M

Table 1

Training data
929,000 tokens
Epochs
6,000
Batch size
70,000

They use BPTT with length 35. During architecture search data batch size is 20 and network batch size is 100. While training the architecture they end up finding, batch size is 64. So effective batch size is 35 * 20 * 100 = 70,000 during architecture search, and 35 * 64 = 2,240 during final training.

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
7.3 × 10¹⁷ FLOP

PTB has 912344 tokens. The model has 23M parameters and was trained for 6k epochs. If the model was dense, 6 FLOP/token/param/epoch * 6k epochs * 23M params * 912k tokens = 1.05e18 FLOP. Alternatively, the model was trained on 1 V100 GPU and "In terms of efficiency, the overall cost... is within 1 GPU day" so the training time was around or below 24 hours. Half precision and 30% utilization would be a pretty good match for the arithmetic estimate: 24 hours * 30% * 28 TFLOPS = 7.3e17 FLOP.

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 V100
Chips used
1
Wall-clock time
24 hours
Power draw
339 W

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

PTB dataset

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 show that an architecture found by WeNets arXiv:1904.03819v1 [cs.NE] 8 Apr 2019 WeNet: Weighted Networks for Recurrent Network Architecture Search achieves state-of-the-art results on the Penn Treebank language dataset"

Record confidence
Likely
Citations
5
Benchmark data
WeNet (WT2)

Sources

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

Reference
WeNet: Weighted Networks for Recurrent Network Architecture Search
Last updated
11 February 2026

What the numbers mean

Where it came from

WeNet (Penn Treebank) was published by Amazon, in United States of America, in April 2019. It comes out of industry.

It works in Language, and is recorded as doing neural Architecture Search - NAS, Language modeling.

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

Training and provenance

Training it took roughly 7.3 × 10¹⁷ FLOP of computation, on NVIDIA V100 — a measure of what producing the model cost, not of how fast it answers.

Around 929,000 tokens went into training it.

Its inclusion criterion is sOTA improvement.

Answers

WeNet (Penn Treebank) — common questions

01

How many parameters does WeNet (Penn Treebank) have?

WeNet (Penn Treebank) has 23M parameters. Table 1. 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 WeNet (Penn Treebank)?

WeNet (Penn Treebank) was published by Amazon, based in United States of America, categorised as industry.

03

When was WeNet (Penn Treebank) released?

WeNet (Penn Treebank) was published in April 2019. 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 WeNet (Penn Treebank) used for?

WeNet (Penn Treebank) works in Language, and is recorded as handling neural Architecture Search - NAS, 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.

05

How much compute was used to train WeNet (Penn Treebank)?

Around 7.3 × 10¹⁷ 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 WeNet (Penn Treebank)?

None. WeNet (Penn Treebank) 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 WeNet (Penn Treebank) open source?

No. WeNet (Penn Treebank) has not had its weights published, so it exists only as a service controlled by its owner.

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.