QRNN
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
- Salesforce Research
- Organisation type
- Industry
- Country
- United States of America
- Published
- 1 February 2018
- Authors
- Stephen Merity, Nitish Shirish Keskar, James Bradbury, Richard Socher
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
- 135M
- Training data
- 103,000,000 tokens
- Epochs
- 14
- Batch size
- 8,400
Based on the details provided in the paper, the number of parameters in the QRNN model can be calculated as follows: Embedding layer size: 400 Number of QRNN layers: 4 Number of nodes per QRNN layer: 2500 Vocabulary size: 267,735 (for WikiText-103 dataset) Adaptive softmax layer Parameters: Embedding layer: 400 x 267,735 = 107,094,000 QRNN layers: Input to hidden weights per layer: 400 x 2500 = 1,000,000 Hidden to hidden weights per layer: 2500 x 2500 = 6,250,000 Biases per layer: 2500 = 2,500…
"We train with a batch size of 60 and a sequence length of 140." ""Using this approach we reduce our per-epoch time substantially and achieve a new state-of-the-art on WikiText-103 despite training for 14 epochs, a total time of only 12 hours."
60*140
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
- 6.9 × 10¹⁷ FLOP
- How it was established
- Operation counting,Hardware
6 FLOP / parameter / token * 135000000 parameters ['Likely' confidence] * 103000000 tokens * 14 epochs = 1.16802e+18 FLOP 31330000000000 FLOP / sec / GPU [fp16 assumed] * 1 GPU * 12 hours * 3600 sec / hour * 0.3 [assumed utilization] = 4.060368e+17 FLOP sqrt(4.060368e+17*1.16802e+18) = 6.8866472e+17 FLOP
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
- Wall-clock time
- 12 hours
"The model trains at 2980 seconds per epoch on the NVIDIA V100 and 5460 seconds per epoch on the NVIDIA P100." "Using this approach we reduce our per-epoch time substantially and achieve a new state-of-the-art on WikiText-103 despite training for 14 epochs, a total time of only 12 hours."
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
- Record confidence
- Likely
- Citations
- 5
- Benchmark data
- QRNN
"we reduce our per-epoch time substantially and achieve a new state-of-the-art on WikiText-103 despite training for 14 epochs"
Sources
Where this record came from and when it was last checked.
- Reference
- Scalable Language Modeling: WikiText-103 on a Single GPU in 12 hours
- Last updated
- 1 December 2025
What the numbers mean
What this model is
QRNN was published by Salesforce Research, in United States of America, in February 2018. industry is the category the publisher falls under.
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.
Training and provenance
Producing it required around 6.9 × 10¹⁷ FLOP of arithmetic, on NVIDIA V100, which is a statement about the training budget rather than about inference.
It was trained on about 103,000,000 tokens of text.
It is tracked in the underlying dataset for one reason in particular: sOTA improvement.
Answers
QRNN — common questions
What is QRNN used for?
QRNN 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.
How much compute was used to train QRNN?
Around 6.9 × 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.
What GPU do I need to run QRNN?
None. QRNN 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.
Is QRNN open source?
No. QRNN has not had its weights published, so it exists only as a service controlled by its owner.
How many parameters does QRNN have?
QRNN has 135M parameters. Based on the details provided in the paper, the number of parameters in the QRNN model can be calculated as follows: Embedding layer size: 400 Number of QRNN layers: 4 Number of nodes per QRNN layer: 2500 Vocabulary size: 267,735 (for WikiText-103 dataset) Adaptive softmax layer Parameters: Embedding layer: 400 x 267,735 = 107,094,000 QRNN layers: Input to hidden weights per layer: 400 x 2500 = 1,000,000 Hidden to hidden weights per layer: 2500 x 2500 = 6,250,000 Biases per layer: 2500 = 2,500 Total QRNN layers parameters: 4 x (1,000,000 + 6,250,000 + 2,500) = 28,000,000 Adaptive softmax layer: No extra parameters due to weight tying Total Parameters = Embedding + QRNN layers = 107,094,000 + 28,000,000 = 135,094,000 So the total number of parameters in the QRNN model is approximately 135 million. The majority of parameters are in the embedding layer, while the 4 QRNN layers contribute 28 million parameters. The adaptive softmax does not add any extra parameters due to weight tying. 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.
Who created QRNN?
QRNN was published by Salesforce Research, based in United States of America, categorised as industry.
When was QRNN released?
QRNN was published in February 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.
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.