Skip to content

Pull requests: Tencent/hpc-ops

New pull request New

16 Open 49 Closed

Pull requests list

[Fused MoE] Extend W13 with peer-indexed EP input

#89 opened Aug 25, 2026 by taoyuanyuan

Loading...

chore: unify remaining copyright headers to Tencent 2026

#88 opened Aug 25, 2026 by xueyangcs Contributor

Loading...

fix: move pipeline barriers into dynamic SharedStorage for 128-byte alignment (#42)

#87 opened Aug 24, 2026 by XFDG

Loading...

2

# docs: add layout algebra comment to make_tiled_copy_V_interleave_trans (#44)

#86 opened Aug 24, 2026 by XFDG

Loading...

fix: fix kTaskLoopPolicy type in grouped gemm host code

#85 opened Aug 23, 2026 by icyhearts

Loading...

2

fix(fuse_moe): correct dowm_tmas typo

#83 opened Aug 19, 2026 by cherishhh

Loading...

Improve source build robustness

#78 opened Jul 31, 2026 by liuyao0322

Loading...

Fix AllReduce benchmark README script name

#77 opened Jul 31, 2026 by liuyao0322

Loading...

[ROCm] Add ROCm support and norm op.

#69 opened Jul 20, 2026 by WhatGhost

Loading...

1

perf: add small-batch route MMA for MoE

#68 opened Jul 20, 2026 by Religious-J Contributor

Loading...

feat: add qwen3 tts rope operator

#67 opened Jul 17, 2026 by l-wave

Loading...

fix: change kHalfFltEps from 1e-20 to 5e-8 in sampler rng, aligned to torch

#65 opened Jul 13, 2026 by LINZHI97

Loading...

feat: add qwen3 tts topk gumbel sampler

#64 opened Jul 10, 2026 by l-wave

Loading...

feat: relax heads_per_group constraint to support 1~8 for BF16 and FP8 decode attention

#58 opened Jun 25, 2026 by Religious-J Contributor

Loading...

feat: bf16 fused moe kernel

#41 opened Apr 9, 2026 by ZelinMa557

Loading...

5

feat: add bf16 group gemm kernel

#34 opened Mar 23, 2026 by ZelinMa557

Loading...

ProTip! Follow long discussions with comments:>50.