Skip to content

Pull requests: flagos-ai/FlagGems-vllm

New pull request New
40 Open 112 Closed
40 Open 112 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

[Operator][XCS] Add a TLE OP swiglu on Ascend
#728 opened Aug 30, 2026 by neoskypig Collaborator Loading…
3 tasks
1
[MUSA] Optimize deepseek_v4 fused_q_kv_rmsnorm for Moore Threads
#726 opened Aug 28, 2026 by yy33min Loading…
3 tasks
fix(enflame): avoid grid.x overflow in scaled_int8_quant static path
#725 opened Aug 28, 2026 by douxetpur Collaborator Loading…
3 tasks
1
[KMCompiler] Fix persistent_topk stale ring slots and robust vLLM tes…
#724 opened Aug 28, 2026 by LittleShun1214 Contributor Loading…
1 of 3 tasks
[KMCompiler]top_k_per_row_prefill&top_k_per_row_decode: adopt FlagGems' suite, fix two vendor-blocking defects, add an MTT prefill override
#723 opened Aug 28, 2026 by cheersluvs • Draft
[AdvancedCompiler]optimize fused add rms norm
#722 opened Aug 27, 2026 by zhoubo567 Loading…
3 tasks
1
optimize mhc
#708 opened Aug 26, 2026 by huangyiqun Collaborator • Draft
3 tasks
[TLE ]Add TLE kernel for topk_softplus_sqrt (perf optimization)
#705 opened Aug 24, 2026 by ph0375 Loading…
3 tasks
1
Optimize moe_sum with a pair-load fast path
#704 opened Aug 24, 2026 by laylowch Loading…
3 tasks
1
Optimize deepseek_v4_attention_fused_q_kv_rmsnorm
#703 opened Aug 24, 2026 by yy33min Loading…
3 tasks
1
Add int8/fp8 quantized sparse attention operator ci/nvidia Opt a fork PR into the H20 nvidia tests lane
#702 opened Aug 24, 2026 by yysheng26 Contributor Loading…
1 of 3 tasks
[Operator] Add Hopper FP8 einsum with persistent TLE barrier ring
#699 opened Aug 19, 2026 by Prophet-Hongyi Loading…
1 of 3 tasks
1
[TLE] Optimize fp8_fp4_paged_mqa_logits with a TLE TMA+WGMMA fast path
#698 opened Aug 18, 2026 by henghengxiedaima Loading…
3 tasks
1
[TLE] Optimize fp8_fp4_mqa_logits with a TLE WGMMA fast path
#697 opened Aug 18, 2026 by henghengxiedaima Loading…
3 tasks
[KMCompiler]Fused moe: add thead (PPU-ZW810E) backend
#696 opened Aug 18, 2026 by LittleShun1214 Contributor Loading…
1 of 3 tasks
feat: optimize w8a8 block fp8 matmul for Hopper
#693 opened Aug 18, 2026 by zikong20 Loading…
3 tasks
1
migrate flaggems.modules.*
#688 opened Aug 17, 2026 by huangyiqun Collaborator Loading…
3 tasks
[KMCompiler]Fused MoE: sync generic ops with FlagGems upstream and converge platform work into vendor-specific backends ci/benchmark ci/nvidia Opt a fork PR into the H20 nvidia tests lane vendor/MetaX vendor/MooreThreads
#687 opened Aug 15, 2026 by LittleShun1214 Contributor Loading…
1 of 3 tasks
1
[KMCompiler][MetaX][Hygon][Mthreads][T-Head][Ascend] fused_deepseek_v4_qnorm_rope_kv_rope_quant_insert backends, plus test and benchmark fixes
#684 opened Aug 14, 2026 by cheersluvs • Draft
1
[KMCompiler]Fix fused moe vendor gating (metax fast paths)
#682 opened Aug 14, 2026 by LittleShun1214 Contributor Loading…
1 of 3 tasks
1
fix: sync FlagGems run_tests resilience fixes
#680 opened Aug 14, 2026 by wuwentao Contributor Loading…
1
[KMCompiler] perf-opt grouped_topk for DeepSeek-v3.2
#92 opened Aug 7, 2026 by liuxiao0909c • Draft
1
feat(ascend): add mhc_pre_clamp_sinkhorn_backward operator for Ascend 910C
#91 opened Aug 7, 2026 by chx7514 Contributor Loading…
1 of 3 tasks
1
feat(ascend): add mhc_pre_clamp_sinkhorn operator for Ascend 910C
#89 opened Aug 7, 2026 by chx7514 Contributor Loading…
1 of 3 tasks
feat(ascend): add mhc_post_backward operator for Ascend 910C
#88 opened Aug 7, 2026 by chx7514 Contributor Loading…
1 of 3 tasks
ProTip! Mix and match filters to narrow down what you’re looking for.