v0.0.16

What's Changed

• chore: cp313 by @UranusSeven in #101
• fix: allow configurable grpc memssage size by @Kevin-XiongC in #103
• refactor(sglang): remove pipeline parallelism support from radix cache by @jimmy-evo in #104
• perf(save): optimize save path with batched GPU copies and hash-first insert by @xiaguan in #106
• feat(connector): add MLA support for vLLM KV connector by @xiaguan in #107
• refactor(trace): replace cfg boilerplate with trace helper macros and add sampling by @xiaguan in #108
• Add MetaServer for multi-node block hash coordination by @jimmy-evo in #88
• fix(connector): use actual layer count for DSA models by @wz1qqx in #112
• test(core): add block_key_copy benchmark by @jimmy-evo in #114
• chore: extract python-binding and server-ops skills from CLAUDE.md by @xiaguan in #115
• fix(python): restore gRPC client message size limits by @wz1qqx in #116
• feat(core): support SSD alignment padding for unaligned block sizes by @xiaguan in #117
• fix(core): per-slot NUMA affinity for SSD prefetch allocation by @xiaguan in #119
• feat: pegaflow-transfer RDMA path with Mooncake-aligned perf by @xiaguan in #102
• fix(core): fix inflight bytes overcount and move insert worker to std::thread by @xiaguan in #122
• refactor(core): tighten visibility and remove dead code by @xiaguan in #125
• test(core): add GPU integration tests for PegaEngine by @xiaguan in #127
• feat(connector): add DCP/PCP context-parallel support for vLLM KV connector by @Kevin-XiongC in #126
• feat(connector): add cross-layer block support by @xiaguan in #128
• feat(connector): enable cross-layer blocks by default by @xiaguan in #129
• chore: bump version to 0.0.15 by @xiaguan in #133
• chore: bump version to 0.0.16 by @xiaguan in #134

New Contributors

• @UranusSeven made their first contribution in #101
• @Kevin-XiongC made their first contribution in #103

Full Changelog: v0.0.15...v0.0.16