All issues

Search results

[Performance] SM100: packed varlen L<64 is 4.4x slower than padded L64 (guarded tail staging vs TMA)

Status: Open.

#43 In QwenLM/FlashQLA;

· SakiNana7qi opened on Sep 1, 2026

FP16 forward pass blows up with NaNs past 32k context length

Status: Open.

#38 In QwenLM/FlashQLA;

· kds1123001 opened on Aug 24, 2026

[Bug] SM120 forward: 4 tests fail on varlen configs

Status: Open.

#35 In QwenLM/FlashQLA;

· LeoHugh opened on Aug 9, 2026

Findings from integrating FlashQLA's Blackwell GDN kernels into a JAX trainer

Status: Open.

#23 In QwenLM/FlashQLA;

· celiolarcher opened on Jul 3, 2026

FlashQLA GDN backend produces different tokens from FLA Triton on correctness-test prompts

Status: Open.

#15 In QwenLM/FlashQLA;

· JessePinkman2019 opened on May 11, 2026

[feature request] decode kernel & mtp kernel

Status: Open.

#11 In QwenLM/FlashQLA;

· ZelinMa557 opened on May 2, 2026

Support for Blackwell

enhancementNew feature or requestNew feature or request

Status: Open.

#2 In QwenLM/FlashQLA;

· mzusman opened on Apr 29, 2026