Issues

• [NVIDIA] low_latency_all_to_all_v2: int32 buffer offsets overflow at large num_experts * max_m * hidden, corrupting dispatch/combine

  ByteDance-Seed/Triton-distributed#203 · Filter by author zkyue opened on Aug 29, 2026

• copy_tensor: the persistent 2-D arm computes row offsets in int32, so a copy whose row extent passes 2 GiB reads and writes out of bounds

  ByteDance-Seed/Triton-distributed#201 · Filter by author truong-v opened on Aug 25, 2026

• gqa_fwd_batch_decode: every non-power-of-two head dim fails to compile — the rope-split Q load indexes the wrong axis

  ByteDance-Seed/Triton-distributed#200 · Filter by author truong-v opened on Aug 25, 2026

• Pinned Ruff check reports 21 errors in AMD modules

  ByteDance-Seed/Triton-distributed#199 · Filter by author dajiaohuang opened on Aug 24, 2026

• Native license check fails for 39 tracked source files

  ByteDance-Seed/Triton-distributed#198 · Filter by author dajiaohuang opened on Aug 24, 2026

• Strict Sphinx build has duplicate objects and unresolved PyTorch references

  ByteDance-Seed/Triton-distributed#195 · Filter by author dajiaohuang opened on Aug 24, 2026

• Kernel API reference signatures are out of sync with implementations

  ByteDance-Seed/Triton-distributed#194 · Filter by author dajiaohuang opened on Aug 24, 2026

• setup.py fails to recognize repository checkouts

  ByteDance-Seed/Triton-distributed#192 · Filter by author dajiaohuang opened on Aug 24, 2026

• Build guide references nonexistent .codebase test scripts

  ByteDance-Seed/Triton-distributed#190 · Filter by author dajiaohuang opened on Aug 24, 2026

• LittleKernel Scalar Types table is missing from rendered docs

  ByteDance-Seed/Triton-distributed#188 · Filter by author dajiaohuang opened on Aug 24, 2026

• gen_amd_libdevice_extra.py --out cannot accept a file path

  ByteDance-Seed/Triton-distributed#186 · Filter by author dajiaohuang opened on Aug 24, 2026

• [BUG] reduce_1d_persistent_kernel: tl.atomic_add into a torch.empty

  ByteDance-Seed/Triton-distributed#182 · Filter by author truong-v opened on Jul 29, 2026

• calc_gather_scatter_index_v2_kernel: unmasked histogram reduction counts padding lanes into expert 0 when ntokens * topk % 1024 != 0

  ByteDance-Seed/Triton-distributed#181 · Filter by author truong-v opened on Jul 21, 2026

• swiglu_backward writes every row's grad_fc1_output to row 0 (missing dA/dB output row-pointer offset) → NaN backward in fused EP-MoE

  ByteDance-Seed/Triton-distributed#180 · Filter by author tianxiao-rgb opened on Jul 17, 2026

• Are there any plans to support nvshmem4py-cu13?

  ByteDance-Seed/Triton-distributed#177 · Filter by author binbinzhm opened on Jun 15, 2026

• [feature] Implement of UniEP?

  ByteDance-Seed/Triton-distributed#176 · Filter by author LiJunscs opened on Jun 8, 2026

• Possible incorrect barrier offset in tile_kernel_dispatch_token_intra_node zero-token expert path

  ByteDance-Seed/Triton-distributed#169 · Filter by author oliverYoung2001 opened on May 7, 2026

• [BUG] Multi-node reduce-scatter crash due to unregistered output buffer in NVSHMEM P2P

  ByteDance-Seed/Triton-distributed#155 · Filter by author Gracia-tjw opened on Feb 3, 2026
  2

• [Bug] symm_at fails with constexpr_type when buffer parameter is None - "'constexpr_type' object has no attribute 'is_block'"

  ByteDance-Seed/Triton-distributed#152 · Filter by author ziyuhuang123 opened on Jan 21, 2026
  1

• Why some tile task does NOT set InputDependencyDesc property?

  ByteDance-Seed/Triton-distributed#141 · Filter by author swordfate opened on Dec 5, 2025
  1

  

• TileLink release plans

  ByteDance-Seed/Triton-distributed#135 · Filter by author gueraf opened on Nov 10, 2025
  

• Bug:TP=4 init_nvshmem_by_torch_process_group errors with PAI-Megatron-Patch on 8 X A100

  ByteDance-Seed/Triton-distributed#132 · Filter by author xuel0707 opened on Oct 23, 2025
  2

• Bug: assert_allclose fails in test_ag_moe.py for large model shape (K=14336) due to numerical divergence

  enhancement
  ByteDance-Seed/Triton-distributed#127 · Filter by author ziyuhuang123 opened on Sep 24, 2025
  1

  

• Overlap is slower than non-overlap in MoE+RS on A100*8

  enhancement
  ByteDance-Seed/Triton-distributed#125 · Filter by author Xqh142857 opened on Sep 22, 2025
  2

• All2all performance comparison between Triton-distributed and DeepEP

  bug
  ByteDance-Seed/Triton-distributed#117 · Filter by author Wangyaoyuu opened on Sep 17, 2025
  5