We reverse-engineered Flash Attention 4 (modal.com) | HN Screenshots

YHacker Newsscreenshots

Screenshot of modal.com: We reverse-engineered Flash Attention 4

First screen, captured 1 year ago.

Looks similar

Screenshot of modal.com: GPU memory snapshots: sub-second startup (2025)

GPU memory snapshots: sub-second startup (2025)(modal.com)

16 points by jxmorris12 9 months ago | 5 comments

Screenshot of modal.com: Boosting multimodal inference performance by >10% with a single Python dict

Boosting multimodal inference performance by >10% with a single Python dict (modal.com)

10 points by jxmorris12 5 months ago | 0 comments

Screenshot of modal.com: Lambda on hard mode: serverless HTTP in Rust

Lambda on hard mode: serverless HTTP in Rust (modal.com)

114 points by pierremenard 2 years ago | 35 comments

Screenshot of modal.com: Modal Auto Endpoints: Optimized inference you own

Modal Auto Endpoints: Optimized inference you own (modal.com)

11 points by handfuloflight 4 months ago | 0 comments

Screenshot of pytorch.org: FlexAttention: The Flexibility of PyTorch with the Performance of FlashAttention

FlexAttention: The Flexibility of PyTorch with the Performance of FlashAttention (pytorch.org)

194 points by limoce 2 years ago | 21 comments

Screenshot of together.ai: FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-Precision

FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-Precision (together.ai)

269 points by jhshah 2 years ago | 53 comments