Morph MORPH
SIGN IN SIGN UP
dhruvhead / FlexKV UNCLAIMED
0 0 0 Python

COMMITS

main
July 30, 2026
X
Improve FlexKV reliability and operational logging (#233)
Xing Liu committed
5186a7c
July 29, 2026
Z
Bugfix (#244)
ZHUOFAN LI committed
010efd2
Z
feat(transfer): chunked cudaHostRegister via FLEXKV_CUDAHOST_CHUNK_SIZE_GB (#240)
ZHUOFAN LI committed
39bf0ae
P
Add graceful shutdown for FlexKV for avoiding cpu buffer leak. (#238)
Psycho Xu committed
951e89b
L
fix(transfer): propagate transfer failures instead of silently dropping them (#237)
laoshuyao committed
3487146
July 28, 2026
L
fix(kvtask): roll back plan resources when cancelling a never-launched task (#236)
laoshuyao committed
a9ebf53
July 27, 2026
L
perf(transfer): dispatch from a dirty set instead of sweeping every graph (#235)
laoshuyao committed
bf28ea5
July 24, 2026
P
Merge/dpsk4 mooncake store (#231)
Psycho Xu committed
3ecd83f
衔
fix(kvtask): release tasks after early response and graph completion (#230)
衔尾 committed
120f307
July 23, 2026
衔
fix: avoid unnecessary CPU allocation for GDS GET (#229)
衔尾 committed
0f8acc2
July 22, 2026
Z
feat(ce): parallel CPU gather/scatter with NT store (#227)
zitto committed
6c2b4ef
July 21, 2026
X
feat(dsv4): add DeepSeek-V4 KV cache and layerwise transfer support (#225)
Xing Liu committed
56c71a7
July 17, 2026
Z
CE transfer: adaptive five-path KV-cache strategy with ping-pong staging (#203)
zitto committed
b70205e
July 8, 2026
S
test(batch_kvtask): drop stale dp_id kwarg to match post-PP-refactor API (#197)
Stary committed
3c968fc
July 7, 2026
H
feat(nvfp4): NVFP4 KV cache support for FlexKV + vLLM integration (#204)
Hao Xu committed
f527d1e
S
perf(transfer): fold KV dim into inner loop + CE block merging + benchmark integration (#198)
Stary committed
5be1ebe
Z
feat: add layerwise polling notification mode + transfer bugfixes (#199)
zitto committed
16174c8
July 3, 2026
S
perf(transfer): merge consecutive blocks in CE path to reduce cudaMemcpyAsync calls (#193)
Stary committed
e83df97
Z
feat: MLA D2H transfer modes with comprehensive tests (#192)
zitto committed
7e1a305
June 30, 2026
衔
fix: support vLLM 0.23+ non-MLA KV layout via new LAYERBLOCK type (#187)
衔尾 committed
d0bcc0e
June 24, 2026
衔
refactor: own sub-tasks at batch merge to fix batch task lifecycle leak (#184)
衔尾 committed
aa74e39
H
Fix reversed CPU block order in H2REMOTE transfer during global PUT (#186)
hu_hamster committed
c346477
L
feat(nvcomp): integrate nvcomp ANS compression and assorted fixes
linhu-nv committed
01389bc
L
feat(pp): add Pipeline Parallel support with WorkerKey/RankInfo refactor
linhu-nv committed
1a91e7f
L
feat(hugepage): add HugePage support
linhu-nv committed
2c5fd53
L
feat(layerwise): layerwise KV transfer + sglang/mooncake integration
linhu-nv committed
71dfa69
June 16, 2026
衔
fix: cancel KVTask immediately when match/put condition not met to prevent memory leak (#176)
衔尾 committed
90e653c
June 9, 2026
Z
support cpu_only match (#179)
ZHUOFAN LI committed
8b4f95a
June 8, 2026
H
perf: avoid per-IO iovec allocation in SSD io_uring path (#177)
Hao Xu committed
db1880c
June 4, 2026
H
Docs inclusive eviction (#175)
Hao Xu committed
ae9b35a