NVIDIA/recsys-examples v26.04

v26.04

Repository: NVIDIA/recsys-examples

Tag: v26.04

Published: 2026-05-20T06:35:18Z

Prerelease: no

Release notes:

What's Changed

Features & Enhancements

• Add hash_roundrobin routing mode to mitigate modulo-aliasing imbalance by @ShaobinChen-AH in https://github.com/NVIDIA/recsys-examples/pull/367
• Jagged Arbitrary Masked Self Attention support by @z52527 in https://github.com/NVIDIA/recsys-examples/pull/339
• fix segmented_unique_cuda: replace table_ids with segmented_range by @jiashuy in https://github.com/NVIDIA/recsys-examples/pull/377
• perf(hstu): restore eager-mode .item() in preprocessor; drop duplicate triton_jagged.py by @JacoCheung in https://github.com/NVIDIA/recsys-examples/pull/389
• Improve AOTI compilation of hstu model by @geoffreyQiu in https://github.com/NVIDIA/recsys-examples/pull/380
• Recsys KVCache Manager refactored into standalone package by @geoffreyQiu in https://github.com/NVIDIA/recsys-examples/pull/387
• Add inference aoti benchmark results by @geoffreyQiu in https://github.com/NVIDIA/recsys-examples/pull/394
• [FEA] Beam search by @z52527 in https://github.com/NVIDIA/recsys-examples/pull/379

Bug Fixes

• fix: unify dense tensor padding convention (dim-0 == batch_size) by @JacoCheung in https://github.com/NVIDIA/recsys-examples/pull/362
• fix(dynamicemb): traverse nn.Module children in check_emb_collection_modules by @JacoCheung in https://github.com/NVIDIA/recsys-examples/pull/355
• fix(ddp): bucket_size=True silently disables grad bucketing by @JacoCheung in https://github.com/NVIDIA/recsys-examples/pull/374

Misc

• fix: reduce Docker layers, add auto CI trigger, fix fake ops import by @JacoCheung in https://github.com/NVIDIA/recsys-examples/pull/363
• delete invalid line by @ShaobinChen-AH in https://github.com/NVIDIA/recsys-examples/pull/381
• build(docker): bump Megatron-LM 0.12.1 -> 0.13.1 to fix count_zeros wasted work by @JacoCheung in https://github.com/NVIDIA/recsys-examples/pull/375
• Clean up unused variables in get_kvcache_metadata_buffer by @gameofdimension in https://github.com/NVIDIA/recsys-examples/pull/371
• Update blossom-ci.yml by @z52527 in https://github.com/NVIDIA/recsys-examples/pull/391

Full Changelog: https://github.com/NVIDIA/recsys-examples/compare/v26.03...v26.04