Skip to content

Pull requests: flashinfer-ai/flashinfer

New pull request New
596 Open 3,135 Closed
596 Open 3,135 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

feat: enable FP8 large-head attention on SM80 op: attention
#5044 opened Sep 9, 2026 by jhaotingc Contributor • Draft
11 tasks
1
perf(prefill): reduce with Tensor.max() instead of the Python builtin in plan() op: attention
#5043 opened Sep 9, 2026 by gf239 Loading…
5 tasks done
1
feat(mla): add CUDA graph plan update API op: attention
#5041 opened Sep 8, 2026 by saltyminty Collaborator Loading…
5 of 11 tasks
3
refactor(kda): stop 0.7 advertising unreleased KDA surface (mark prefill wrapper experimental, trim kda_kernels.__all__) op: linear attention KDA, mamba, GDN, etc. review filtering.
#5040 opened Sep 8, 2026 by kahyunnam Member Loading…
10 tasks done
3
fix(attention): default skip_all_rows_active_check to true op: attention run-ci
#5039 opened Sep 8, 2026 by saltyminty Collaborator Loading…
3 of 11 tasks
1
4
fix(decode): reserve max(K+V, st.o) smem for FA2 FP8+GQA decode op: attention run-ci
#5038 opened Sep 8, 2026 by ir1ka Contributor Loading…
5 of 11 tasks
6
fix(kda): make recurrent_kda backend="auto" decode fall back to CuTe DSL op: linear attention KDA, mamba, GDN, etc. review filtering. run-ci v0.6.19
#5037 opened Sep 8, 2026 by kahyunnam Member Loading…
5 tasks done
1
11
Add SM100 NVFP4 warp decode and standalone SiLU op: moe run-ci
#5036 opened Sep 8, 2026 by yyihuang Collaborator • Draft
3
test(kda): restore the frozen module-ident consistency check op: linear attention KDA, mamba, GDN, etc. review filtering. run-ci
#5035 opened Sep 8, 2026 by kahyunnam Member Loading…
5 tasks done
2
fix(sampling): make multi-CTA top-k renorm deterministic (fixed-order cross-CTA sum) op: misc norm, activation, sampling, RoPE, quantization, etc.
#5034 opened Sep 8, 2026 by gilfordting • Draft
3 tasks done
1
Add TOPK=256 to the SM120 NVFP4 sparse-MLA dispatch envelope op: attention
#5033 opened Sep 8, 2026 by danielwoz Loading…
1
feat(kda): add CuTe DSL small-BH KDA prefill backend for Blackwell op: linear attention KDA, mamba, GDN, etc. review filtering.
#5032 opened Sep 8, 2026 by qiangxu1996 Loading…
5 of 11 tasks
2
fix(moe): Fix per-token quantization fast-math INF scale when activation contains all-zero row op: moe
#5031 opened Sep 8, 2026 by xuantengh Contributor Loading…
4 of 5 tasks
6
fix(attention): initialize variable-length attention sum accumulator op: attention
#5029 opened Sep 8, 2026 by shuo-ouyang Loading…
5 of 11 tasks
1
fix(cake_kda): fix C16 forward triangular matrix orientation op: linear attention KDA, mamba, GDN, etc. review filtering.
#5028 opened Sep 8, 2026 by zheyang0825 Loading…
4 of 11 tasks
1
feat(comm): add allocation-stable head-chunk Ulysses primitives op: comm
#5027 opened Sep 8, 2026 by tiffany940107 Contributor Loading…
2 of 11 tasks
12
fix(moe): keep per-token FP8 quant scales finite for tiny activations op: moe
#5024 opened Sep 8, 2026 by yilin-void Loading…
6 tasks done
1
feat(comm): add BF16 PCIe IPC all-gather and reduce-scatter op: comm
#5023 opened Sep 8, 2026 by yilin-void • Draft
1
Support compact GLM NoPE FP8 rows and fix masked reads op: attention
#5022 opened Sep 8, 2026 by ormandj Contributor Loading…
1 of 11 tasks
1
feat(kda): support graph-safe CuTe DSL prefix checkpoints op: linear attention KDA, mamba, GDN, etc. review filtering. run-ci
#5021 opened Sep 8, 2026 by Observer007 Contributor Loading…
5 of 11 tasks
7
fix(attention): pre-scale P by 2^8 before e4m3 quantization in SM120 prims FP8 prefill op: attention
#5020 opened Sep 8, 2026 by wqyg18 Loading…
5 tasks done
1
feat(moe): add CuTe DSL NVFP4 W4A16 MegaMoE
#5019 opened Sep 8, 2026 by zianglih Contributor • Draft
4 of 11 tasks
1
test(attention): exercise the NVFP4 split-KV gated path on hardware op: attention
#5018 opened Sep 7, 2026 by Champollion9012 Loading…
5 of 11 tasks
1
fix(ci): preserve estimate file compression during refresh
#5014 opened Sep 7, 2026 by mottopanikeiku Loading…
4 of 11 tasks
2
fix(activation): pick a vector width that divides hidden_size in act_and_mul (fixes d=3420) op: misc norm, activation, sampling, RoPE, quantization, etc.
#5013 opened Sep 7, 2026 by Wint3rNight Loading…
5 tasks done
1
1
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.