Open Models Almanac
Methodsmodel architecturetoken mixersoftmax attention

specific method · filed under model architecture

Scaled Dot-Product Attention (SDPA)

Computes attention using scaled dot products and softmax; the cited implementations use PyTorch SDPA, including with custom masks.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See softmax attention for the methods filed beside it.

sources
2
model
1
lab adopt it
1
strongest
used

Documented in

Allen Institute for AI

Evidence

2 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
used 2

MolmoAct 2

Allen Institute for AIadopted2 sources

use PyTorch’s scaled dot-product attention implementation rather than FlashAttention because our packed multimodal sequences and robot-conditioning masks require custom attention masks.

usedunclearin MolmoAct2 training implementationAllen Institute for Artificial Intelligence (AI2)
1 more from this model

use PyTorch’s Scaled Dot-Product Attention (SDPA) attention implementation.

usedsoftware implementationin MolmoActAllen Institute for Artificial Intelligence (AI2)
↑