Skip to content

SYSTEM Cited by 1 source

TransAct (Pinterest)

Definition

TransAct is Pinterest's sequential user-action Transformer model — named alongside the Pinterest Foundation Model as a primary consumer of the ~16K-token user action sequences that dominate Pinterest's request-level data (Source: sources/2026-04-13-pinterest-scaling-recommendation-systems-with-request-level-deduplication).

Published at arXiv 2506.02267 (not ingested on the wiki; referenced by the 2026-04-13 post as the long user sequence modeling component alongside the Foundation Model).

Why it's wiki-relevant

TransAct appears in the 2026-04-13 request-level-deduplication post as one of the canonical consumers of user sequences that drove deduplication-as-a-discipline:

"Request-level data is massive. It largely consists of user sequences, approximately 16K tokens encoding all actions a user has taken on the platform. These sequences power sequential user understanding components like the [Pinterest Foundation Model] and [TransAct]."

TransAct predates the Foundation Model and operates on the same ~16K-token sequence substrate; both benefit equally from request-level dedup (storage, training, serving).

Caveats

  • Stub — this post only mentions TransAct by name; architectural details come from the arXiv paper, not the 2026-04-13 post.
  • No production scale, training-throughput, or serving-throughput numbers disclosed in the ingested source.
  • Relationship to Foundation Model (chronological? architectural? shared training stack?) not disclosed in the 2026-04-13 post.
  • Specific downstream applications (Home Feed ranking? Search ranking? Ads?) not itemised.

Seen in

Last updated · 550 distilled / 1,221 read