Meta Pseudo Labels
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 Brain,Google AI
- Organisation type
- Industry,Industry
- Country
- United States of America
- Published
- 1 March 2021
- Authors
- Hieu Pham, Zihang Dai, Qizhe Xie, Minh-Thang Luong, and Quoc V. Le
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Vision
- Task
- Image classification
- 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
- 480M
- Training data
- 131,280,000 tokens
Table 4 480M
Section 4 Datasets. For this experiment, we use the entire ImageNet training set as labeled data, and use the JFT dataset as unlabeled data. The JFT dataset has 300 million images, and then is filtered down to 130 million images by Noisy Student using confidence thresholds and up-sampling [77]. We use the same 130 million images as Noisy Student
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
- 4.8 × 10²² FLOP
- How it was established
- Hardware
From communication with author: 22671 TPU days on specific hardware. Which hardware did you use and in which configuration? 2048 cores of TPU v3. Precision: Mixed. bfloat16 for activations, float32 for weights and optimizer slots. 2048 TPUv3 cores means 1024 TPUv3 chips, and the spec is 123e12 FLOP/second per chip with bfloat16 precision (Source: https://cloud.google.com/tpu/docs/system-architecture-tpu-vm) So the compute estimate is: 1024 chips * 123e12 FLOP/second * 0.4 utilization * 11 d…
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
- Google TPU v3
- Chips used
- 1,024
- Chip-hours
- 270,336
- Wall-clock time
- 264 hours (11 days)
- Power draw
- 934.4 kW
- Compute cost
- $53,844
11 days from section 4: "We train the model for 1 million steps in total, which takes about 11 days for EfficientNet-L2 and 10 days for EfficientNet-B6-Wide. " "Specifically, our training process runs on a cluster of 2,048 TPUv3 cores. "
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
- Open source
Apache-2.0 license https://github.com/google-research/google-research/blob/master/meta_pseudo_labels/README.md
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
- 766
"We present Meta Pseudo Labels, a semi-supervised learning method that achieves a new state-of-the-art top-1 accuracy of 90.2% on ImageNet, which is 1.6% better than the existing state-of-the-art."
Sources
Where this record came from and when it was last checked.
- Reference
- Meta pseudo labels
- Last updated
- 25 May 2026
What the numbers mean
Where it came from
Meta Pseudo Labels was published by Google Brain,Google AI, in United States of America, in March 2021. The organisation is categorised as industry,Industry.
It works in Vision, and is recorded as doing image classification.
Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.
Training and provenance
Producing it required around 4.8 × 10²² FLOP of arithmetic, on Google TPU v3, which is a statement about the training budget rather than about inference.
The training set ran to roughly 131,280,000 tokens.
It is tracked in the underlying dataset for one reason in particular: sOTA improvement.
Answers
Meta Pseudo Labels — common questions
How many parameters does Meta Pseudo Labels have?
Meta Pseudo Labels has 480M parameters. Table 4 480M. 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 Meta Pseudo Labels?
Meta Pseudo Labels was published by Google Brain,Google AI, based in United States of America, categorised as industry,Industry.
When was Meta Pseudo Labels released?
Meta Pseudo Labels was published in March 2021. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
What is Meta Pseudo Labels used for?
Meta Pseudo Labels works in Vision, and is recorded as handling image classification. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
How much compute was used to train Meta Pseudo Labels?
Around 4.8 × 10²² FLOP, on Google TPU v3. 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 Meta Pseudo Labels?
None. Meta Pseudo Labels 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 Meta Pseudo Labels open source?
No. Meta Pseudo Labels has not had its weights published, so it exists only as a service controlled by its owner.
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.