Skip to content

Pull requests: flashinfer-ai/flashinfer

New pull request New
403 Open 2,555 Closed
403 Open 2,555 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

chore: add ROCm/aiter submodule
#4076 opened Jul 20, 2026 by demandal25 Loading…
2 tasks done
1
feat(moe_ep): close the nixl_ep transport gaps blocking the vLLM Fleet/Handle adapters
#4075 opened Jul 20, 2026 by Anerudhan Collaborator • Draft
5
docs(moe_ep): add NIXL-EP test instructions to the runbook
#4074 opened Jul 20, 2026 by Anerudhan Collaborator • Draft
3
perf(jit): prune arch gencode that dispatch can never load op: gemm run-ci
#4073 opened Jul 20, 2026 by yichengj0 Contributor Loading…
2
feat: add optional normed-output (y_out) to add_rmsnorm_fp4quant run-ci
#4072 opened Jul 20, 2026 by nv-guomingz Loading…
5 tasks done
13
feat(experimental): add flashinfer.experimental.sm12x
#4071 opened Jul 20, 2026 by lukealonso Contributor Loading…
5 tasks done
29
feat(moe_ep): add SM90 push FP8 mega-MoE backend for Hopper op: comm op: gemm
#4069 opened Jul 20, 2026 by leonardHONG Contributor Loading…
8 of 9 tasks
13
perf(moe): avoid copying batched GEMM configs run-ci
#4068 opened Jul 20, 2026 by DongyunZou Loading…
5 tasks done
1
perf(moe): phase compact EP4 NVFP4 dispatch payload copies
#4067 opened Jul 20, 2026 by samnordmann Contributor • Draft
1
fix: avoid SM120 MXFP8 MXFP4 autotune crash op: moe
#4066 opened Jul 20, 2026 by waynehacking8 Contributor Loading…
1
1
perf(sm103): optimize NVFP4 epilogue stores and scaling op: gemm run-ci
#4063 opened Jul 20, 2026 by tiffany940107 Contributor Loading…
4 of 5 tasks
10
perf(triton): single-load fast path for plain RMSNorm
#4060 opened Jul 20, 2026 by yoavkantor Loading…
7
fix: uniform multi-token trtllm-gen decode mis-indexes packed Q/O without cum_seq_lens_q (#3929) op: attention
#4059 opened Jul 20, 2026 by thanhhao98 Contributor Loading…
2 tasks done
1
6
docs: fix MSA and GDN doc-check findings run-ci v0.6.16
#4058 opened Jul 20, 2026 by kangbintNV Collaborator Loading…
6
feat(moe): support caller-owned CUTLASS MoE workspace op: moe
#4057 opened Jul 20, 2026 by feih-nv Collaborator Loading…
3 tasks done
7
fix(norm): clamp CUDA FP8 quantization to the output dtype range
#4056 opened Jul 19, 2026 by morluto Loading…
1
5
fix(gemm): validate tensorwide bmm_fp8 scales op: gemm
#4055 opened Jul 19, 2026 by morluto Loading…
1
4
fix(cascade): preserve empty attention states during merges op: attention
#4054 opened Jul 19, 2026 by morluto Loading…
1
7
Fix XQA strided query handling op: attention
#4050 opened Jul 19, 2026 by mmangkad Loading…
5 tasks
6
feat(moe): support BF16 activations in CuTe DSL NVFP4 MoE op: moe
#4048 opened Jul 18, 2026 by zianglih Contributor • Draft
5 tasks done
7
perf(sampling): reduce top-p target probability memory
#4047 opened Jul 18, 2026 by qianlihuang Loading…
5 tasks done
7
fix(gemm): prevent out-of-bounds read in MaskedScheduler op: gemm
#4046 opened Jul 18, 2026 by dashanji Loading…
4 of 5 tasks
1
fix(fused_moe): avoid repeated runner setup in host dispatch op: moe
#4045 opened Jul 18, 2026 by YAMY1234 Contributor Loading…
9
Sf only per tok quant op: moe
#4042 opened Jul 17, 2026 by IwakuraRein Collaborator • Draft
5 tasks
5
Revert "[Bugfix] Skip autotuning for small token counts on SM100 to prevent TMA crash" op: moe run-ci
#4041 opened Jul 17, 2026 by b8zhong Contributor Loading…
@qiching
9
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.