Pull requests · meta-recsys/generative-recommenders

Reland: rename `silu_u` to `concat_u_silu_u` in `STUNormPostProcConfig`

#590melody-lui3 opened

Sep 25, 2026

1

jagged_dense_bmm_broadcast_add CuTeDSL Kernel

#589keegs13d opened

Sep 24, 2026

Contributor

1

Prune costly wide RMSNorm autotune configs on SM100

#588tangyuhao opened

Sep 24, 2026

Contributor

1

Fix int32 overflow in _helion_split_2d_jagged_kernel indexing

#584rafaykhurram opened

Sep 23, 2026

Contributor

2

Re-land: Add RMSNorm as an option for both HSTU STU norms (training)

#582Sidharth123-cpu opened

Sep 16, 2026

1

Cross-attn inference fwd: use the 128-row tile when it fills the GPU

#581Sidharth123-cpu opened

Sep 16, 2026

1

fused layernorm kernels in hstu v2

#580ycw66 opened

Sep 15, 2026

2

Add RMSNorm as an option for both HSTU STU norms (training)

#579Sidharth123-cpu opened

Sep 14, 2026

2

Add an RMSNorm branch to the fused HSTU norm kernels

#578Sidharth123-cpu opened

Sep 11, 2026

1

Let pytorch_rms_norm accept weight=None

#577code-by-jin opened

Sep 2, 2026

1

Fix dropout mask fake output for B200

#574ruochen99 opened

Aug 24, 2026

Contributor

1

optimize jagged_jagged_reduce_sum for MoT

#573ruochen99 opened

Aug 21, 2026

Contributor

1

Enable AOTI tracing for symbolic HSTU shapes

#572ruochen99 opened

Aug 14, 2026

Contributor

1

Avoid unused optional tensor allocations

#571njriasan opened

Aug 10, 2026

Contributor

1

Suppress new Pyrefly `constexpr` errors in generative_recommenders Triton ops

#570njriasan opened

Aug 10, 2026

Contributor

1

Swap `-c` CUDA version flags for the constraint modifier

#569wenxin0319 opened

Aug 7, 2026

Contributor

1

num_stages=0 -> 1 (meta-recsys/generative-recommenders)

#568njriasan opened

Aug 4, 2026

Contributor

1

Fix num_stages range

#567njriasan opened

Aug 4, 2026

Contributor

1

Fix Triton layer norm u activation forwarding

#566wang1ang opened

Aug 4, 2026

Use one compiler stage for H100 TLX configs

#564mark14wu opened

Jul 31, 2026

2

Add local attention kernels and coverage

#562njriasan opened

Jul 30, 2026

Contributor

1

Fix: use bitwise operators instead of logical operators in Triton kernels

#561wang1ang opened

Jul 28, 2026

Make torchrec.modules.sorted_index_select import defensive

#560LinjianMa opened

Jul 24, 2026

Contributor

1

HSTU CA-overarch ranker architecture, config providers, and kernels

#558XialuLi opened

Jul 22, 2026

1

Optimize jagged-jagged BMM reduce-sum kernel

#557arsatis opened

Jul 21, 2026

1

Navigation Menu

All pull requests

Reland: rename `silu_u` to `concat_u_silu_u` in `STUNormPostProcConfig`

jagged_dense_bmm_broadcast_add CuTeDSL Kernel

Prune costly wide RMSNorm autotune configs on SM100

Fix int32 overflow in _helion_split_2d_jagged_kernel indexing

Re-land: Add RMSNorm as an option for both HSTU STU norms (training)

Cross-attn inference fwd: use the 128-row tile when it fills the GPU

fused layernorm kernels in hstu v2

Add RMSNorm as an option for both HSTU STU norms (training)

Add an RMSNorm branch to the fused HSTU norm kernels

Let pytorch_rms_norm accept weight=None

Fix dropout mask fake output for B200

optimize jagged_jagged_reduce_sum for MoT

Enable AOTI tracing for symbolic HSTU shapes

Avoid unused optional tensor allocations

Suppress new Pyrefly `constexpr` errors in generative_recommenders Triton ops

Swap `-c` CUDA version flags for the constraint modifier

num_stages=0 -> 1 (meta-recsys/generative-recommenders)

Fix num_stages range

Fix Triton layer norm u activation forwarding

Use one compiler stage for H100 TLX configs

Add local attention kernels and coverage

Fix: use bitwise operators instead of logical operators in Triton kernels

Make torchrec.modules.sorted_index_select import defensive

HSTU CA-overarch ranker architecture, config providers, and kernels

Optimize jagged-jagged BMM reduce-sum kernel