GitHub · Where software is built

Labels Milestones New issue

Mixed-version deployments fail with QueryPrefetch UNIMPLEMENTED; add version/capability reporting

#136

· xiaguan opened

on Mar 9, 2026

KV-cache registration is O(layers) and gets expensive for large models

good first issue

#132

· xiaguan opened

on Mar 8, 2026

CUDA graph compatibility: layer-wise KV ops require PIECEWISE mode

#124

· xiaguan opened

on Mar 3, 2026

bug(core): readv/writev may exceed UIO_MAXIOV (1024) iovec limit on large TP configurations

#120

· xiaguan opened

on Mar 2, 2026

Add custom io_uring benchmark (harness=false) and align with fio

good first issue

#100

· xiaguan opened

on Feb 6, 2026

Add Regular (cached) pinned memory allocation strategy

good first issue

#95

· xiaguan opened

on Feb 4, 2026

[Feature] NUMA-aware cache eviction to prevent cross-node thrashing

#85

· xiaguan opened

on Feb 2, 2026

Revisit E2E Fuzzing Test Strategy Due to Minor Response Variance

#45

· xiaguan opened

on Jan 21, 2026

[CI/CD] CI tests for pegaflow-server

#40

· wz1qqx opened

on Jan 19, 2026

End-to-End Accuracy Test for Pega KV Cache

good first issue

#34

· xiaguan opened

on Jan 16, 2026

feat(vllm): Add prefetch_threshold Parameter

good first issue

#18

· xiaguan opened

on Jan 12, 2026

[perf]: Improve Memory Utilization of Offset Allocator

#11

· xiaguan opened

on Jan 8, 2026