Project AURIONAI Use-Cases Research & Innovation
recommendation

HSTU (Generative Recommenders)

Meta generative recommender; up to 65.8% NDCG over baselines, 5.3x to 15.2x faster than FlashAttention2 transformers at 8,192 length, +12.4% online A/B. Code Apache-2.0.

MetaMetaUSA
Meta
Levelmodel
Classrecommendation
Parameters1.5T
Contextnot in the notes
Released27 Feb 2024
Opennessopen source
Runs onhyperscale cluster
Statusactive

Lineage

successor
HSTU (Generative Recommenders) to MTGR: successor, medium confidenceHSTU (Generative Recommenders)HSTU (Generative Recommend…Meta, Feb 2024MTGRMTGRMeituan, May 2025
MeituanMTGR is a child (successor)medium confidencenotes
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 2 models. Confidence and evidence are the notes' own.

R&D Topics

Generative recommendationexemplar

Identity

Developer
Meta
Family
not in the notes
Released
27 Feb 2024 sourcesannouncement page
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
dense transformer Hierarchical Sequential Transduction Unit; recommendation as sequential transduction
Total parameters
1.5T sources
Active parameters
–
Checkpoint parameters
–
Note
Deployed models up to 1.5 trillion parameters
Context length
not in the notes
Input
action
Output
action
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
open source
Licence
not in the notes sources
Access routes
internal only
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
hyperscale cluster
Footprint
not in the notes
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Benchmark Results

The notes record no benchmark result for this model.

Industry Use Cases

No industry use case links this model.

Sources

  1. Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations arxiv.org https://arxiv.org/abs/2402.17152
  2. GitHub - meta-recsys/generative-recommenders: Repository hosting code for "Actions Speak Louder than Words: Trillion-Parameter Sequential Tr github.com/facebookresearch/generative-recommenders https://github.com/facebookresearch/generative-recommenders
  3. GitHub - meta-recsys/generative-recommenders: Repository hosting code for "Actions Speak Louder than Words: Trillion-Parameter Sequential Tr github.com/meta-recsys/generative-recommenders https://github.com/meta-recsys/generative-recommenders

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "HSTU (Generative Recommenders)", https://aurion.semai.id/models/meta-hstu/, records of October 2026.