JFT
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
- Google Research,Carnegie Mellon University (CMU)
- Organisation type
- Industry,Academia
- Country
- United States of America
- Published
- 10 July 2017
- Authors
- Chen Sun, Abhinav Shrivastava, Saurabh Singh, Abhinav Gupta
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Vision
- Task
- Image classification, Object detection, Semantic segmentation, Pose estimation
- 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
- 44.7M
- Training data
- 5,487,300,000,000 tokens
- Epochs
- 4
- Batch size
- 32
Uses ResNet-101 architecture, which has 44,654,504 parameters: https://resources.wolframcloud.com/NeuralNetRepository/resources/ResNet-101-Trained-on-ImageNet-Competition-Data/
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
- 8.4 × 10²⁰ FLOP
- How it was established
- Hardware
Tesla K80 performance: 8.13 TFLOP/s Assume 40% utilization 60 days * 50 GPUs * 40% utilization * 8.13 TFLOP/s/GPU = 8.43*10^20 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 Tesla K80
- Chips used
- 50
- Chip-hours
- 72,000
- Wall-clock time
- 1,440 hours (60 days)
- Power draw
- 31.3 kW
- Compute cost
- $17,911
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Frontier model
- Yes
- Why it is tracked
- SOTA improvement
- Record confidence
- Confident
- Citations
- 2,712
SOTA on COCO and PASCAL VOC
Sources
Where this record came from and when it was last checked.
- Reference
- Revisiting Unreasonable Effectiveness of Data in Deep Learning Era.
- Last updated
- 25 May 2026
What the numbers mean
Where it came from
JFT was published by Google Research,Carnegie Mellon University (CMU), in the country recorded as United States of America, during July 2017. The publishing organisation is categorised as industry,Academia.
It works in the domain of Vision, and is recorded as performing the task of image classification, Object detection, Semantic segmentation, Pose estimation.
This is a closed model: the trained values stayed with whoever produced them, and there is no local version to run.
How it was trained
The training run consumed about 8.4 × 10²⁰ FLOP, on hardware recorded as NVIDIA Tesla K80. That figure measures what producing the model cost, and has no bearing on how fast it answers.
It was trained on a corpus of about 5,487,300,000,000 tokens of text.
Its inclusion criterion: sOTA improvement.
Answers
JFT — common questions
JFT— how many parameters does it have?
It has a parameter count of 44.7M. Uses ResNet-101 architecture, which has 44,654,504 parameters: https://resources.wolframcloud.com/NeuralNetRepository/resources/ResNet-101-Trained-on-ImageNet-Competition-Data/. 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.
JFT— who created it?
It was published by Google Research,Carnegie Mellon University (CMU), based in United States of America, an organisation categorised as industry,Academia.
JFT— when was it released?
It was published in July 2017. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
JFT— what is it used for?
It works in the domain of Vision, and is recorded as handling the task of image classification, Object detection, Semantic segmentation, Pose estimation. These are the areas it was designed around; they describe intent rather than a hard boundary.
JFT— how much compute was used to train it?
Training consumed around 8.4 × 10²⁰ FLOP, on hardware recorded as NVIDIA Tesla K80. 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.
JFT— 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.
JFT— is it open source?
The licensing was never recorded in our source data. We treat unstated licensing as closed, because an unrecorded licence is not one to rely on.
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.