Decaying Fast Weights Transformer (WT-103)
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
- Jenni
- Organisation type
- Industry
- Country
- Singapore
- Published
- 9 October 2022
- Authors
- Huanru Henry Mao
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Language
- Task
- Language modeling
- Base model
- Transformer (Adaptive Input Embeddings) WT103
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
- 242M
- Training data
- 103,000,000 tokens
- Epochs
- 192.12
"We fine-tune starting from the checkpoint provided by Baevski and Auli (2019)8, which has 242M parameters"
"For both models, we trained with a batch size of 26 for 100,000 iterations. "
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
- 7.9 × 10¹⁹ FLOP
- How it was established
- Hardware
- Fine-tuning compute
- 5.7 × 10¹⁸ FLOP
Pre-trained model is Transformer (Adaptive Input Embeddings) which was 7.3e19. This is from 8 * 67 V100-hours. fine-tuning: "Training was performed on a single NVIDIA A40 GPU for 35 hours" 35h, 1 GPU, 149.7e12, 30% = 5.7e18 FLOP" 5.7e18 + 7.3e19 is 7.9e19
"Training was performed on a single NVIDIA A40 GPU for 35 hours" 35h, 1 GPU, 149.7e12, 30% 5.7e18 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 A40 PCIe
- Chips used
- 1
- Wall-clock time
- 35 hours
- Power draw
- 330 W
"Training was performed on a single NVIDIA A40 GPU for 35 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
They have CUDA kernels, don't see pretrain code: https://github.com/jenni-ai/T2FW
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
- 30
- Benchmark data
- Decaying Fast Weights Transformer
Sources
Where this record came from and when it was last checked.
- Reference
- Fine-Tuning Pre-trained Transformers into Decaying Fast Weights
- Last updated
- 25 May 2026
What the numbers mean
Where it came from
Decaying Fast Weights Transformer (WT-103) was published by Jenni, in the country recorded as Singapore, during October 2022. The category the publisher falls under is industry.
It works in the domain of Language, and is recorded as performing the task of language modeling.
Rather than being trained from scratch, it is derived from Transformer (Adaptive Input Embeddings) WT103. Most models at this scale are adapted from an existing base rather than built from nothing.
Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.
How it was trained
Producing it required arithmetic totalling around 7.9 × 10¹⁹ FLOP, on hardware recorded as NVIDIA A40 PCIe. That figure measures what producing the model cost, and has no bearing on how fast it answers.
The training set ran to roughly 103,000,000 tokens of text.
Answers
Decaying Fast Weights Transformer (WT-103) — common questions
Decaying Fast Weights Transformer (WT-103)— what is it used for?
It works in the domain of Language, and is recorded as handling the task of language modeling. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
Decaying Fast Weights Transformer (WT-103)— how much compute was used to train it?
Training consumed around 7.9 × 10¹⁹ FLOP, on hardware recorded as NVIDIA A40 PCIe. 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.
Decaying Fast Weights Transformer (WT-103)— 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.
Decaying Fast Weights Transformer (WT-103)— is it open source?
No. Its weights have not been published, so it exists only as a service controlled by its owner.
Decaying Fast Weights Transformer (WT-103)— how many parameters does it have?
It has a parameter count of 242M. "We fine-tune starting from the checkpoint provided by Baevski and Auli (2019)8, which has 242M parameters". 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.
Decaying Fast Weights Transformer (WT-103)— who created it?
It was published by Jenni, based in Singapore, an organisation categorised as industry.
Decaying Fast Weights Transformer (WT-103)— when was it released?
It was published in October 2022. 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.