-
Notifications
You must be signed in to change notification settings - Fork 579
Pull requests: linkedin/Liger-Kernel
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Pin fp64 scalar kernel params to fp32 in modulated_rms_norm and attn_res
#1364
opened Aug 10, 2026 by
buffett0323
Contributor
Loading…
1 of 3 tasks
Pin fp64 scalar kernel params to fp32 in cross entropy
#1362
opened Aug 10, 2026 by
buffett0323
Contributor
Loading…
3 tasks done
[Liger Cute] Fix fused MoE combine for partial token tiles
#1361
opened Aug 10, 2026 by
supercharleszhu
Contributor
Loading…
3 tasks
Fix global LayerNorm mutation in model patching
#1360
opened Aug 10, 2026 by
vedavyas2727
Loading…
1 of 3 tasks
Pin fp64 scalar kernel params to fp32 in remaining norm ops
#1358
opened Aug 9, 2026 by
buffett0323
Contributor
Loading…
3 tasks done
[Hopper] Add CuTe DSL and cuTile fused linear cross entropy
#1356
opened Aug 8, 2026 by
Celaena24
Contributor
Loading…
Fix ZeroDivisionError in fused linear cross entropy when weight vocab dim is 0
#1353
opened Aug 7, 2026 by
nileshpatil6
Loading…
1 of 3 tasks
[Perf] Fuse log-softmax into fused_linear_jsd kernel
#1352
opened Aug 7, 2026 by
hiwuhgds-pixel
Loading…
3 tasks done
fix(rope): don't mark the unused batch-size param as tl.constexpr
#1351
opened Aug 7, 2026 by
vaibhavjindal
Collaborator
•
Draft
1 of 3 tasks
[NPU] align fused linear JSD forward with accum_dtype API
#1346
opened Aug 6, 2026 by
zheliuyu
Contributor
Loading…
3 tasks done
Cast tl.program_id(0) to int64 in llama4_rope and qwen2vl_mrope
#1343
opened Aug 5, 2026 by
akshansh47
Loading…
1 task
Cast tl.program_id to int64 in llama4_rope and qwen2vl_mrope
#1342
opened Aug 5, 2026 by
adityasingh2400
Loading…
1 of 3 tasks
[NPU] Fix: recompute true-class CE gradient in fp32 on no-weight backward path
#1341
opened Aug 5, 2026 by
zheliuyu
Contributor
Loading…
3 tasks done
fix: torch.compile loss corruption at ignore_index rows (+ perf: drop BT x V loss buffer in FLJSD)
#1338
opened Aug 4, 2026 by
KyleMylonakisProtopia
Contributor
Loading…
3 tasks done
perf(cutedsl): fused dX+dW backward + host-enqueue caching for RMSNorm
#1336
opened Aug 3, 2026 by
Charlesrizzly
Contributor
Loading…
Fix/device context for remaining kernels
#1331
opened Aug 2, 2026 by
kashif
Contributor
Loading…
3 tasks done
Fix FLCE torch.compile failure from aten.addmm.dtype_out overload
#1328
opened Jul 31, 2026 by
yashb98
Loading…
2 of 3 tasks
Add Megatron-Core integration for RoPE
#1325
opened Jul 27, 2026 by
PrathyushaPolepalli
Contributor
•
Draft
3 tasks
[Perf] Use direct FLCE weight-gradient accumulation for FP16, BF16, and FP32
#1324
opened Jul 25, 2026 by
Anubhav-2003
Loading…
7 of 9 tasks
Previous Next
ProTip!
Exclude everything labeled
bug with -label:bug.