[TRTLLM-12499][feat] Add support for chunked KVCache transfer for disaggregated serving in Python Cache Transceiver - #15519

Closed

athena-nv wants to merge 11 commits into

NVIDIA:mainfrom

athena-nv:trtllm-11608-chunked-kvcache-transfer-combined

Commits on Jul 1, 2026

[TRTLLM-11608][feat] Chunked KV cache transfer with early block release

chienchunhung
authored and
athena-nv
committed
1448c64
fix: preserve mamba_state_index in chunked slices and fix VSWA shared counter

chienchunhung
authored and
athena-nv
committed
0f9034c
Address CodeRabbit review nitpicks

chienchunhung
authored and
athena-nv
committed
764fc81
Address reviewer comments from eopXD and pcastonguay

chienchunhung
authored and
athena-nv
committed
e72926d
Move chunk_block_offset into KVSlice dataclass

chienchunhung
authored and
athena-nv
committed
13f4a88
fix ups
athena-nv
committed
c73dbf9
address coderabbit comments
athena-nv
committed
eb39a17
Revert early block release implementation
athena-nv
committed
d463e35
Project chunks onto resident block ID lists to support chunked transfer of assymetric layer groups
athena-nv
committed
46940a4
Update chunked KV cache transfer tests
athena-nv
committed
4d73fcf
Rename chunk_size_blocks to transfer_chunk_size
athena-nv
committed
2b6e803