Laplace 3B Linear Attention Model Achieves 120k Tokens/s Training and 11,544 Tokens/s Inference on AWS Trainium 2
Current overview (not historical)
The Laplace 3B linear‑attention model now trains at roughly 120,000 tokens per second and infers at about 11,544 output tokens per second on AWS Trainium 2 using native PyTorch. The key innovation is a recursive state that remains a constant 29.5 MB per sequence, independent of context length, and is only 1/82 the size of a traditional KV cache. This compact state enables efficient scaling to long‑context workloads while keeping memory usage low. The benchmark demonstrates that the model can be trained and served on the same hardware platform, highlighting the practicality of linear‑attention architectures for large‑scale natural language processing tasks on specialized AI accelerators. The report also notes that the throughput figures are achieved without specialized optimizations beyond the native framework support, suggesting further gains are possible with additional tuning.
Written with AI from the reporting · updated 10/20/2026
EventGraph Explorer
One time cutoff for reports, relations and evidence.
New at this step ·i14 Trains and Serves Laplace 3B Linear‑Attention Model on AWS Trainium2
- Reports in replay
- 1
- First-party
- 1
- Sources
- 1
Reports first seen at this step · 1
- AWS Machine Learning Blog · First-partyi14 Trains and Serves Laplace 3B Linear‑Attention Model on AWS Trainium2Original source ↗
Discovery-time replay; the linked graph and evidence share this cutoff. The full overview and activity remain current.
Event developments
1 developments10/21/2026, 5:39:12 AM
i14 Trains and Serves Laplace 3B Linear‑Attention Model on AWS Trainium2AWS Machine Learning Blog
Filtered by first discovery; only representative reports already discovered by this cutoff are shown. Story grouping may have changed later, so this is not a complete historical snapshot.
Relations at this time
Loading time-scoped relations…
Bounded reconstruction, not an exact snapshot: entity names, relationship text and story grouping may include later revisions. Overview and full source archive below remain current.
Story route
A station-by-station route from the first public evidence onward; every station carries its own evidence and uncertainty.
- 1Oct 21, 00:00recorded Oct 21, 05:39single source
在AWS Trainium2上训练并服务i14的Laplace 3B线性注意力模型
Read original source ↗ AWS Machine Learning BlogReplay this moment in the explorer
Uncertainty marks flag where to check the sources; they are not proven causation.
Full source archive (current)
Current public reports; outside the historical replay.
First-party sources · 1 reports
- AWS Machine Learning Blog · Oct 21, 05:14i14 Trains and Serves Laplace 3B Linear‑Attention Model on AWS Trainium2
i14’s Laplace 3B, a 3‑billion‑parameter linear‑attention language model, was trained on a single trn2.48xlarge AWS Trainium2 instance at roughly 120,000 tokens per second. During inference, the model delivered 11,544 output tokens per second across 1,280 concurrent sequences while keeping a fixed 29.5 MB recurrent state per sequence, independent of context length. The work leveraged TorchNeuron Native to port i14’s Gated Linear Attention (GLA) kernels to Trainium2, preserving the model’s Muon optimizer and custom architecture.
Read original source ↗
Independent coverage · 0 reports
No verifiable reports in this group yet.
Different reports do not necessarily disagree. Check original sources and publication dates.
Report timeline
Explore every side through dated source reports.
- AWS Machine Learning BlogSelectedi14 Trains and Serves Laplace 3B Linear‑Attention Model on AWS Trainium2
i14’s Laplace 3B, a 3‑billion‑parameter linear‑attention language model, was trained on a single trn2.48xlarge AWS Trainium2 instance at roughly 120,000 tokens per second. During inference, the model delivered 11,544 output tokens per second across 1,280 concurrent sequences while keeping a fixed 29.5 MB recurrent state per sequence, independent of context length. The work leveraged TorchNeuron Native to port i14’s Gated Linear Attention (GLA) kernels to Trainium2, preserving the model’s Muon optimizer and custom architecture.