SILC-S* (86M)
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
- ETH Zurich,DeepMind,Google,Technical University of Munich
- Organisation type
- Academia,Industry,Industry,Academia
- Country
- Switzerland, United Kingdom of Great Britain and Northern Ireland, United States of America, Germany
- Published
- 20 October 2023
- Authors
- Muhammad Ferjad Naeem, Yongqin Xian, Xiaohua Zhai, Lukas Hoyer, Luc Van Gool, Federico Tombari
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Vision
- Task
- Image classification, Image segmentation
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
- 86M
- Training data
- tokens
- Batch size
- 16,000
"SILC models set a new state-of-the-art for these tasks at ViT/B16 model size" (https://arxiv.org/pdf/2310.13355, page 5). VIT/B16-224 has 86.6M parameters, and VIT/B16-384 has 86.9M parameters (https://huggingface.co/google/vit-base-patch16-224, https://huggingface.co/google/vit-base-patch16-384), so I will assume SILC-S has 86M parameters.
SILC-S* saw 20B examples of image-text pairs during training (https://www.ecva.net/papers/eccv_2024/papers_ECCV/papers/03093-supp.pdf, page 4).
"We trained with a batch size of 16k on Google TPUs" (https://arxiv.org/pdf/2310.13355, page 5).
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
- 1 × 10²² FLOP
"SILC* at ViT B/16 can be trained on 256 TPUv4 chips meanwhile the B/8 and L/16 models require 512 chips. The training takes around 5 days" (https://www.ecva.net/papers/eccv_2024/papers_ECCV/papers/03093-supp.pdf, page 7). Assuming a 33% utilization rate, Training compute = utilization rate * # of chips used * peak FLOPS / chip * training time = 0.33 * 256 TPUv4 chip * 2.75e14 FLOPS / TPUv4 chip * 5 days * 24 h / day * 3600 s / h ~= 1.004e22 FLOPS
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 v4
- Chips used
- 256
- Wall-clock time
- 120 hours
- Power draw
- 172.8 kW
"SILC* at ViT B/16 can be trained on 256 TPUv4 chips meanwhile the B/8 and L/16 models require 512 chips. The training takes around 5 days" (https://www.ecva.net/papers/eccv_2024/papers_ECCV/papers/03093-supp.pdf, page 7).
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.
- Record confidence
- Confident
Sources
Where this record came from and when it was last checked.
- Reference
- SILC: Improving Vision Language Pretraining with Self-Distillation
- Last updated
- 11 February 2026
What the numbers mean
About this model
SILC-S* (86M) was published by ETH Zurich,DeepMind,Google,Technical University of Munich, in Switzerland, in October 2023. academia,Industry,Industry,Academia is the category the publisher falls under.
It works in Vision, and is recorded as doing image classification, Image segmentation.
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
Producing it required around 1 × 10²² FLOP of arithmetic, on Google TPU v4, which is a statement about the training budget rather than about inference.
Answers
SILC-S* (86M) — common questions
How much compute was used to train SILC-S* (86M)?
Around 1 × 10²² FLOP, on Google TPU v4. 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 SILC-S* (86M)?
None. SILC-S* (86M) 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 SILC-S* (86M) open source?
No. SILC-S* (86M) has not had its weights published, so it exists only as a service controlled by its owner.
How many parameters does SILC-S* (86M) have?
SILC-S* (86M) has 86M parameters. "SILC models set a new state-of-the-art for these tasks at ViT/B16 model size" (https://arxiv.org/pdf/2310.13355, page 5). VIT/B16-224 has 86.6M parameters, and VIT/B16-384 has 86.9M parameters (https://huggingface.co/google/vit-base-patch16-224, https://huggingface.co/google/vit-base-patch16-384), so I will assume SILC-S has 86M 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.
Who created SILC-S* (86M)?
SILC-S* (86M) was published by ETH Zurich,DeepMind,Google,Technical University of Munich, based in Switzerland, categorised as academia,Industry,Industry,Academia.
When was SILC-S* (86M) released?
SILC-S* (86M) was published in October 2023. 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 SILC-S* (86M) used for?
SILC-S* (86M) works in Vision, and is recorded as handling image classification, Image segmentation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
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.