[None][feat] Add NVFP4 as a cold-page KV Cache Compression Method - #18091

Merged

Hudayday merged 39 commits into

NVIDIA:mainfrom

Hudayday:nvfp4-pr17512-latest-e2e-clean-20260817

Sep 1, 2026

Commits on Aug 24, 2026

[None][feat] port NVFP4 cold-page compression to latest KVCM2
Hudayday
committed
5731c07
[None][chore] Clarify KVCC lifecycle and test tiers
Hudayday
committed
6b502cd
[None][fix] Polish NVFP4 cold-page compression
Hudayday
committed
8bb2c3f
Merge remote-tracking branch 'upstream/main' into nvfp4-pr17512-latest-e2e-clean-20260817
Hudayday
committed
fa8a37c
[None][test] Fix compression compatibility test setup
Hudayday
committed
f593902
Merge remote-tracking branch 'upstream/main' into nvfp4-pr17512-latest-e2e-clean-20260817
Hudayday
committed
eb4418c
[None][fix] Tighten KV cache compression admission
Hudayday
committed
2f8e955

Commits on Aug 25, 2026

[None][test] Fix KV cache compression CI routing
Hudayday
committed
8bb2ed4
[None][test] Preserve existing compression test routing
Hudayday
committed
d9fcc3c
[None][refactor] Move NVFP4 cold-page policy to Python
Hudayday
committed
50db0dc
[None][refactor] Split generic cold-page codec backend
Hudayday
committed
1971d2f
[None][test] Rely on KVCM buffer validation
Hudayday
committed
a2e177e

Commits on Aug 26, 2026

Commits on Aug 27, 2026

Commits on Aug 29, 2026

Commits on Aug 30, 2026

Merge branch 'main' of https://github.com/NVIDIA/TensorRT-LLM into nvfp4-pr18091-python-callback-cold-page-20260825
Hudayday
committed
e5d6374

Commits on Aug 31, 2026