GShard (dense)
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
- Organisation type
- Industry
- Country
- United States of America
- Published
- 30 June 2020
- Authors
- Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, Zhifeng Chen
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Language
- Task
- Translation
- Approach
- Self-supervised learning
- 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
- 2.3B
- Training data
- 346,666,666,667 tokens
- Batch size
- 4,000,000
"Our best quality dense single Transformer model (2.3B parameters) achieving ∆BLEU of 6.1, was trained with GPipe [15] on 2048 TPU v3 cores for 6 weeks or total of 235.5 TPU v3 core-years."
"We focus on improving the translation quality (measured in terms of BLEU score [48]) from all 100 languages to English. This resulted in approximately 13 billion training examples to be used for model training" Each example is a sentence pair. Assuming 20 words per sentence and 4/3 tokens per word, that is 13*20*4/3 billion tokens
Table 3, bolded row is best model
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
- Operation counting,Hardware
Trained for a total of 235.5 TPU v3 core-years. Hardware estimate: 235.5 * 365.25 * 24 * 3600 * (1.23e14 / 2) * 0.3 = 1.371e23 Footnote 10 indicates 300k steps and 4M tokens/step -> 1.2T tokens Arithmetic estimate: 6 * 2.3B * 1.2T = 1.656e22 FLOPs Geometric mean: sqrt(1.371e23 * 1.656e22) = 4.765e22
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
- 1,032,192
- Wall-clock time
- 1,008 hours (42 days)
- Power draw
- 939.5 kW
- Compute cost
- $266,200
6 weeks = 1008 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
- Open source
training code is open, Apache: https://github.com/tensorflow/lingvo/tree/master/lingvo/tasks/lm
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
- 1,992
"such a giant model can efficiently be trained on 2048 TPU v3 accelerators in 4 days to achieve far superior quality for translation from 100 languages to English compared to the prior art" I don't see any standard benchmark that they claim SOTA on
Sources
Where this record came from and when it was last checked.
- Reference
- GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding
- Last updated
- 25 May 2026
What the numbers mean
Where it came from
GShard (dense) was published by Google, in the country recorded as United States of America, during June 2020. The publishing organisation is categorised as industry.
It works in the domain of Language, and is recorded as performing the task of translation.
Its weights were never published, so it can only be reached through its provider. No graphics card changes that.
How it was trained
The training run consumed about 4.8 × 10²² FLOP, on hardware recorded as Google TPU v3. That figure measures what producing the model cost, and has no bearing on how fast it answers.
The training set ran to roughly 346,666,666,667 tokens of text.
It is tracked in the underlying dataset for one reason in particular: sOTA improvement.
Answers
GShard (dense) — common questions
GShard (dense)— how many parameters does it have?
It has a parameter count of 2.3B. "Our best quality dense single Transformer model (2.3B parameters) achieving ∆BLEU of 6.1, was trained with GPipe [15] on 2048 TPU v3 cores for 6 weeks or total of 235.5 TPU v3 core-years.". 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.
GShard (dense)— who created it?
It was published by Google, based in United States of America, an organisation categorised as industry.
GShard (dense)— when was it released?
It was published in June 2020. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
GShard (dense)— what is it used for?
It works in the domain of Language, and is recorded as handling the task of translation. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.
GShard (dense)— how much compute was used to train it?
Training consumed around 4.8 × 10²² FLOP, on hardware recorded as 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.
GShard (dense)— 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.
GShard (dense)— is it open source?
No. Its weights have not been 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.