Issues

• LigerPolyNorm returns wrong (finite) outputs in fp16 for plain randn inputs at hidden size >= 4096

  linkedin/Liger-Kernel#1536 · Filter by author jayden1711 opened on Oct 15, 2026

• LigerFusedLinearKLDivLoss (bf16): loss is 1.6x to 270x too large when student is close to teacher (logits rounded to bf16 before the fp32 loss, same cause as #1432)

  linkedin/Liger-Kernel#1535 · Filter by author jayden1711 opened on Oct 15, 2026

• LigerGroupNorm returns all-zero or NaN groups in fp16 for ordinary inputs, and NaN in bf16 for inputs with a mean offset

  linkedin/Liger-Kernel#1534 · Filter by author jayden1711 opened on Oct 15, 2026

• [RFC] Add TileLang Backend Support

  linkedin/Liger-Kernel#1531 · Filter by author xvxuopop opened on Oct 14, 2026
  2

• Out-of-bounds read in _grpo_loss_bwd_kernel

  linkedin/Liger-Kernel#1529 · Filter by author RoelsJonas opened on Oct 13, 2026
  1

• Triton RoPE uses the wrong token stride when seq_len is 1

  linkedin/Liger-Kernel#1526 · Filter by author KimbingNg opened on Oct 10, 2026

• Convergence mismatches on A100: bf16 Qwen3-MoE / Llama4 loss and fp32 Qwen3.5-MoE top-k logprobs

  linkedin/Liger-Kernel#1508 · Filter by author zheliuyu opened on Oct 3, 2026

• [NPU] Track reverting Ascend Atlas 950 workarounds after triton-ascend upstream fixes (#1464)

  linkedin/Liger-Kernel#1481 · Filter by author zheliuyu opened on Sep 21, 2026

• JSD kernel returns NaN for interior beta when a vocabulary entry underflows under both distributions

  linkedin/Liger-Kernel#1453 · Filter by author KyleMylonakisProtopia opened on Sep 10, 2026
  4

• [RFC] Migrate Nvidia CI from Modal to HF Jobs

  linkedin/Liger-Kernel#1393 · Filter by author ParagEkbote opened on Aug 17, 2026

• [RFC] Testing Methodology for FP32 Convergence Tests on Fused SwiGLU/dSwiGLU Kernels

  linkedin/Liger-Kernel#1383 · Filter by author Pearblossom-M opened on Aug 15, 2026
  1

• Add Liger kernels + FLCE token_accuracy plumbing for Muse Glimmer (muse_glimmer)

  linkedin/Liger-Kernel#1378 · Filter by author MegaGimen opened on Aug 13, 2026
  2

  

• [RFC] Full-pipeline SwiGLU MLP with forward/backward co-optimization

  linkedin/Liger-Kernel#1347 · Filter by author Pearblossom-M opened on Aug 6, 2026
  6

• [B300/SM103] Compiled Liger SwiGLU is ~3.1x slower than torch.compile for a Qwen3-shaped BF16 tensor

  linkedin/Liger-Kernel#1330 · Filter by author jessiewei7 opened on Aug 2, 2026
  2

• torch.compile fails with FLCE due to aten.addmm.dtype_out overload

  linkedin/Liger-Kernel#1327 · Filter by author hiwuhgds-pixel opened on Jul 30, 2026

• Reduce FLCE weight-gradient allocation overhead across FP16, BF16, and FP32

  linkedin/Liger-Kernel#1323 · Filter by author Anubhav-2003 opened on Jul 25, 2026

• Support GMPO Chunked loss

  linkedin/Liger-Kernel#1315 · Filter by author chinoll opened on Jul 23, 2026
  1

• Support Gemma 4 Unified (gemma4_unified) — kernel patching + fused linear cross-entropy

  linkedin/Liger-Kernel#1308 · Filter by author lefft opened on Jul 21, 2026

• Triton kernels can launch on the wrong GPU with multi-GPU inputs

  linkedin/Liger-Kernel#1303 · Filter by author kashif opened on Jul 16, 2026
  1

• Fused MoE kernels crash on Qwen3.5-397B (int32 offset overflow)

  linkedin/Liger-Kernel#1296 · Filter by author likejazz opened on Jul 10, 2026
  1

• Add Qwen2.5-Omni support

  linkedin/Liger-Kernel#1295 · Filter by author VietDunghacker opened on Jul 9, 2026
  1

• Feature Request: Add THD / Packed-Sequence Training Support

  linkedin/Liger-Kernel#1281 · Filter by author niyunsheng opened on Jul 1, 2026
  2

• Suggestion: per-kernel roofline/bandwidth-efficiency benchmarks in docs

  linkedin/Liger-Kernel#1272 · Filter by author ragulk143 opened on Jun 27, 2026
  1

• LigerORPOTrainer doesn't work when not using FSDP

  linkedin/Liger-Kernel#1229 · Filter by author Ola-Vish opened on May 19, 2026
  1

• Add fused Modulated RMSNorm for DiT-style scale/shift conditioning

  linkedin/Liger-Kernel#1224 · Filter by author yukiu00 opened on May 13, 2026
  5