NVIDIA/recsys-examples

View on GitHub

Examples for Recommenders - easy to train and deploy on accelerated infrastructure.

Stars: 322
Forks: 83
Open beginner issues: 0
Indexed issues: 41
Avg merge: 3d 3h
Merged PRs (30d): 8

Dominant language: Python
License: No license data
Last GitHub push: Sep 14, 2026
Latest indexed: Sep 19, 2026
Contributing guide: Contributing guide
Code of conduct: Code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

41 open issues indexed

Split key_value_table.py into multiple files to improve maintainability. Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

NVIDIA/recsys-examples#483 · Opened Sep 10, 2026
[PERF][RFC] Explore a shared-memory AOTI + FlexKV integration for Triton Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#475 · Opened Sep 1, 2026
[FEA] Add KV-cache-aware Triton orchestration and move offload/reap off the inference critical path` Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA/recsys-examples#474 · Opened Sep 1, 2026
[FEA] Add CUDA Graph replay to the HSTU KV-cache AOTI compute path` Open

Difficulty 5/5 Over a week Newbie friendliness 38/100

NVIDIA/recsys-examples#473 · Opened Sep 1, 2026
[BUG] FlexKV SSD-to-host transfer bandwidth is lower than expected Open

bug

NVIDIA/recsys-examples#472 · 2 comments · 1 assignee · Opened Sep 1, 2026
[FEA] Overlap SSD-to-host and host-to-device KV-cache transfers in FlexKV Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA/recsys-examples#471 · Opened Sep 1, 2026
[FEA] Support configurable table-fusion indexing and NVE layer types for exportable inference embeddings Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA/recsys-examples#470 · Opened Sep 1, 2026
Development Roadmap(July/Aug/Sep) Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#462 · Opened Aug 10, 2026
[FEA] Optimize online request scheduling for HSTU inference Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#461 · Opened Aug 10, 2026
[FEA] Add an online serving benchmark for HSTU inference Open

Difficulty 5/5 Over a week Newbie friendliness 30/100

NVIDIA/recsys-examples#460 · Opened Aug 10, 2026
[FEA] Resolve key-to-index mapping in DynamicEmb row-wise input_dist Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

NVIDIA/recsys-examples#459 · Opened Aug 10, 2026
[FEA] Add FP8 KV-cache support for HSTU inference Open

feature

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#457 · Opened Aug 10, 2026
[FEA] Support transformer backbone in HSTU inference Open

feature

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#456 · Opened Aug 10, 2026
[FEA] Support target-aware masks in FA Open

feature

Difficulty 5/5 Over a week Newbie friendliness 15/100

NVIDIA/recsys-examples#455 · Opened Aug 10, 2026
[FEA][Dynamicemb] Support DRAM as an internal storage tier and PS as an external Parameter Store tier Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

NVIDIA/recsys-examples#453 · Opened Aug 10, 2026
[FEA] Add Rubin support for HSTU training Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

NVIDIA/recsys-examples#452 · Opened Aug 10, 2026
[QST] Timeline for "native page tables in decode attention" for SID-GR — variable-length online throughput gap Open

question

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#445 · 1 comment · Opened Jul 31, 2026
[QST] Is this level of run-to-run variance expected when training HSTU retrieval on ML-20M? Open

question

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA/recsys-examples#427 · 1 comment · Opened Jun 21, 2026
[FEA] Recsys-FlexKV: support SSD and Layerwise Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA/recsys-examples#422 · Opened Jun 16, 2026
[FEA] online training example Open

Difficulty 5/5 Over a week Newbie friendliness 15/100

NVIDIA/recsys-examples#420 · Opened Jun 16, 2026
[QST] nsys profile fails to generate profiling file on L20 with version 2603 Open

question

NVIDIA/recsys-examples#397 · 1 comment · 1 assignee · Opened May 19, 2026
[QST] Control over gradient communication (AllReduce) and optimizer steps for decoupled DynamicEmb Open

question TBD

NVIDIA/recsys-examples#393 · 3 comments · 1 assignee · Opened May 15, 2026
[FEA] fused hier output dist Open

Difficulty 5/5 Over a week Newbie friendliness 15/100

NVIDIA/recsys-examples#384 · 1 comment · Opened May 11, 2026
[FEA] remove unnecessary keys selection and fuse selection to keys and scores Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

NVIDIA/recsys-examples#357 · Opened Apr 13, 2026
[FEA] Dynamic bucket routing strategy to avoid uneven load between buckets Open

dynamicemb

NVIDIA/recsys-examples#351 · 1 assignee · Opened Apr 10, 2026
[FEA] Use score_strategy uniformly and remove evict_strategy Open

TBD

Difficulty 4/5 3-5 days Newbie friendliness 35/100

NVIDIA/recsys-examples#344 · 1 comment · Opened Apr 7, 2026
[FEA] Slow unbucketize permute operation in SequenceEmbeddingsAllToAll for row-wise sharding Open

dynamicemb enhancement

NVIDIA/recsys-examples#296 · 1 assignee · Opened Feb 6, 2026
[FEA] Customized optimizer support. Open

dynamicemb feature

Difficulty 5/5 Over a week Newbie friendliness 20/100

NVIDIA/recsys-examples#269 · Opened Jan 14, 2026
[QST] Question about the inference result provided Open

question

NVIDIA/recsys-examples#225 · 2 comments · 1 assignee · Opened Nov 11, 2025
[FEA] Enable FSDP Open

TBD

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#213 · Opened Nov 4, 2025
[Enhancement] optimize the IO for dynamicemb load and dump Open

dynamicemb enhancement TBD

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#182 · Opened Oct 11, 2025
[DOC] Request for Stable AUC Results on Public Datasets Open

documentation

Difficulty 4/5 3-5 days Newbie friendliness 40/100

NVIDIA/recsys-examples#147 · 1 comment · Opened Aug 25, 2025
[QST] what the purpose to reimplement torch.unique as UniqueOp Open

question

NVIDIA/recsys-examples#131 · 3 comments · 1 assignee · Opened Aug 6, 2025
[FEA] Non-blocking dynamicemb dumping Open

dynamicemb TBD

Difficulty 5/5 Over a week Newbie friendliness 15/100

NVIDIA/recsys-examples#125 · Opened Aug 4, 2025
[FEA] Use torchrec input_dist in dynamicemb Open

dynamicemb TBD

Difficulty 5/5 Over a week Newbie friendliness 15/100

NVIDIA/recsys-examples#116 · Opened Jul 30, 2025
[FEA] Decouple dynamic embedding sharding planner with torchrec Open

dynamicemb feature

NVIDIA/recsys-examples#107 · 1 assignee · Opened Jul 22, 2025
[FEA] 8bit inference support in dynamicemb Open

dynamicemb

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#61 · Opened Jun 6, 2025
[FEA] activation offloading Open

feature TBD

NVIDIA/recsys-examples#48 · 1 reaction · 1 assignee · Opened May 21, 2025
[FEA] Enable mcore distributed optimizer Open

TBD

NVIDIA/recsys-examples#41 · 1 assignee · Opened May 14, 2025
[FEA] Support distributed checkpointing Open

TBD

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA/recsys-examples#8 · 2 comments · Opened Apr 23, 2025
[FEA] Support HSTU Context parallelism in training Open

feature

NVIDIA/recsys-examples#7 · 1 assignee · Opened Apr 23, 2025

NVIDIA/recsys-examples

Get new issues in your inbox