v0.0.18
What's Changed
- docs: mark P/D disaggregation as experimental by @xiaguan in #168
- chore(lint): add workspace clippy lint config by @xiaguan in #173
- feat(core): add sharded pinned memory pool via --pool-shards by @feifei-111 in #167
- fix(transfer): RDMA CONNECTING state, logging, and QP metrics by @xiaguan in #175
- fix(build): include pegaflow-metaserver-py binary in wheel by @jimmy-evo in #177
- fix(transfer): prefer RoCEv2 GID for cross-machine RDMA by @xiaguan in #179
- fix(python): add missing req_id parameter to query/query_prefetch stubs by @jimmy-evo in #178
- fix(core): prevent RDMA fetch infinite retry on evicted blocks by @xiaguan in #181
- fix(connector): check pending_save_requests in is_empty() to fix stal… by @feifei-111 in #182
- fix(python): preserve RPC method name in EngineRpcClient.call() errors by @LED-0102 in #180
- chore(python): remove unreachable dead code in init.py by @jimmy-evo in #183
- docs(p2p): add P2P RDMA user guide and simplify CLI by @xiaguan in #186
- refactor(core): remove dead Query RPC and PegaEngine PyO3 binding by @xiaguan in #187
- fix(core): switch RDMA remote fetch to prefix-match semantics by @xiaguan in #188
- feat(transfer): show PCIe link speed and width in topology log by @GentleCold in #193
- fix(connector): adapt handle_preemptions to vLLM #34805 interface change by @xiaguan in #190
- fix(python): compat vllm metrics helpers across releases by @xiaguan in #196
- chore(release): bump pegaflow version to 0.18.0 by @xiaguan in #197
New Contributors
- @LED-0102 made their first contribution in #180
- @GentleCold made their first contribution in #193
Full Changelog: v0.0.17...v0.0.18