forked from deepseek-ai/DeepGEMM
-
Notifications
You must be signed in to change notification settings - Fork 49
Pull requests: sgl-project/DeepGEMM
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[SM90] Validate MegaMoE symmetric buffer peer mappings
#79
opened Aug 25, 2026 by
Andrew88Lin
Loading…
[SM120] Avoid per-token-count JIT recompilation in swapped GEMM paths
#76
opened Aug 14, 2026 by
ormandj
Loading…
fix(layout): validate SF tensors are CUDA; add MXFP4 (1, 32) SF transform test
#75
opened Aug 14, 2026 by
qqtang-code
Loading…
Fix SM90 MegaMoE .template disambiguation on nvcc13
#74
opened Aug 12, 2026 by
Helicoplacoidea
Loading…
2 tasks done
[SM90] Optimize FP8 MegaMoE: fused shared expert, interleaved scheduler, L2 epilogue, 4WG heuristic and decode swap-AB
#69
opened Aug 2, 2026 by
Helicoplacoidea
Loading…
fix(sm90): sanitize pruned FP8 MegaMoE weight blocks
#68
opened Aug 1, 2026 by
ercaoeatting
Loading…
Expose get_best_gemm_config_key for JIT kernel deduplication
#58
opened Jul 10, 2026 by
Dimignat
Loading…
SM90 (Hopper) FP4 MegaMoE fused kernel with swapAB small-batch path
#53
opened Jun 29, 2026 by
qiushixiaoyu
Loading…
[sm100] Packed FP4×FP4 mega-MoE kernel (W4A4) with per-band BLOCK_K
#44
opened Jun 14, 2026 by
Romaosir
Loading…
Fix UE8M0 packing: mask mantissa bits when extracting fp32 exponents
#35
opened May 19, 2026 by
yhyang201
Loading…
ProTip!
Adding no:label will show everything without a label.