Repository navigation

All pull requests

feat: iluvatar argmax
#1040wooway777 opened Nov 13, 2026Contributor19 tasks
feat(qy): QY ops backend with FA and longctx MoE/RoPE kernels (2026.11.12)
#1038Ceng23333 opened Nov 12, 2026Contributor5 of 6 tasks
3
feat: add CUDA QK-norm ops and MLA paged attention for 021 model
#1035rubik-hua opened Nov 11, 2026
22
feat(metax): add MoE, GDN, quantization and attention operators for Metax
#1033rubik-hua opened Nov 10, 2026
feat: add native matmul providers
#1031baominghelly opened Nov 10, 2026Collaborator8 of 20 tasks
feat(cambricon): extend native model operator coverage
#1027hot-strip371 opened Nov 5, 20263 tasks done
1
perf(iluvatar): batch FlashAttnWithKvcache QK reductions
#1026superme1one opened Nov 5, 20263 tasks done
12
perf(iluvatar): fuse causal Softmax with FP32 accumulation
#1025superme1one opened Nov 5, 20267 tasks done
6
perf(iluvatar): generalize small-batch BF16 MLP GEMV
#1024superme1one opened Nov 5, 20262 tasks done
8
feat: add ATen operator implementations
#1022zhoujf1 opened Nov 4, 2026
19
feat: add NineToothed fused QK norm RoPE
#1019voltjia opened Nov 2, 2026Contributor
feat(hygon): add native Clover diffusion and sparse MoE operators
#1015Ringssss opened Oct 29, 20267 tasks done
fix(cpu): reject unsafe Cast storage before writing
#1011superme1one opened Oct 27, 20262 tasks done
perf(iluvatar): migrate validated R5 kernel layouts
#1010superme1one opened Oct 27, 20264 tasks done
1
refactor(aten): unify external tensor wrapping and execution contexts
#1008whjthu opened Oct 25, 2026Contributor19 tasks
feat: support NineToothed RMSNorm on Ascend
#1004voltjia opened Oct 19, 2026Contributor7 tasks done
feat(qy): QY ops backend with FA and longctx MoE/RoPE kernels
#1002Ceng23333 opened Oct 14, 2026Contributor
3
feat: add optimized Kunlun rotary embedding backend
#1001JoeZhang-0x000 opened Oct 10, 20264 tasks done
perf(ascend): accelerate packed Q/K and strided KV/SwiGLU kernels
#1000JoeZhang-0x000 opened Oct 9, 20265 of 8 tasks
perf(metax): reduce RMSNorm blocks for small head dimensions
#999JoeZhang-0x000 opened Oct 9, 2026
perf(metax): vectorize small fused add RMSNorm rows
#998JoeZhang-0x000 opened Oct 9, 2026
perf(metax): vectorize contiguous embedding lookups
#997JoeZhang-0x000 opened Oct 9, 2026
feat: add LightningAttentionInfinilm for MiniMax linear attention
#995y258dd opened Sep 19, 20265 of 19 tasks
feat(metax): add moe_topk_softmax operator for MoE top-k gating
#993rubik-hua opened Sep 18, 2026
feat(metax): add moe_fused_dense operator for fused MoE expert forward
#992rubik-hua opened Sep 18, 2026

Pull requests