Issues

• E2E support for Qwen3 dense; fused QK-RMSNorm + RoPE kernel

  NVIDIA/TileGym#210 · Filter by author nayanp38 opened on Oct 2, 2026

• [ALERT] Access to NVIDIA's Self-Hosted Runners is Expiring

  nvidia-runners
  NVIDIA/TileGym#209 · Filter by author nvidia-runner-mgmt opened on Oct 1, 2026

• Forward Convolution Kernel

  NVIDIA/TileGym#195 · Filter by author m9ssah opened on Sep 4, 2026
  2

• attention_mask is discarded in fmha_interface_wrapper: is a varlen or padding-mask prefill kernel wanted?

  NVIDIA/TileGym#191 · Filter by author Aryan95614 opened on Aug 25, 2026
  1

• Would cuTile ports of Liger's chunked preference losses (DPO, ORPO, SimPO, KTO) be accepted into the liger suite?

  NVIDIA/TileGym#190 · Filter by author Aryan95614 opened on Aug 25, 2026
  1

• Why cutile's rmsnorm required full register

  NVIDIA/TileGym#169 · Filter by author MARD1NO opened on Jul 13, 2026
  5

• Add Gemma 4 text-only inference support to TileGym transformers examples

  NVIDIA/TileGym#130 · Filter by author duoan opened on May 21, 2026
  1

• Training Examples?

  NVIDIA/TileGym#104 · Filter by author vgoklani opened on Apr 9, 2026
  1

• Potential numerical accuracy issue in the fused_moe implementation

  NVIDIA/TileGym#23 · Filter by author huanghua1994 opened on Dec 29, 2025