Releases: IBM/spnl
Releases · IBM/spnl
v0.22.1
v0.22.0
What's Changed
- fix(bench): resolve runtime panic and clap CSV arg parsing; add tests by @starpit in #912
- chore(deps): update actions/attest-build-provenance action to v4 - autoclosed by @renovate[bot] in #914
- chore: update Homebrew formula to v0.21.0 by @github-actions[bot] in #913
- feat: restore (but simpler) chat builtin by @starpit in #916
- chore(deps): update github artifact actions (major) by @renovate[bot] in #915
- refactor: split spnl monolith into 5 crates under crates/ layout by @starpit in #919
- fix: spnl SSE support was bogus by @starpit in #929
- fix: make llir structs pub by @starpit in #928
- chore: bump cli to 0.22.0 by @starpit in #930
Full Changelog: v0.21.0...v0.22.0
v0.21.0
What's Changed
- chore(deps): update rust crate clap to v4.5.60 by @renovate[bot] in #892
- chore: update Homebrew formula to v0.20.1 by @github-actions[bot] in #893
- chore(deps): update rust crate anyhow to v1.0.102 by @renovate[bot] in #894
- fix(ci): use stable cache key for CUDA flash-attn builds by @starpit in #895
- test: add reuse distance measurements to ragcsv benchmark by @starpit in #896
- feat(rag): port augment module from lancedb to leann-rs by @starpit in #897
- chore(deps): bump leann-core 0.1.1 -> 0.1.3, ndarray 0.16 -> 0.17 by @starpit in #899
- chore(deps): bump leann-core 0.1.3 -> 0.1.4 by @starpit in #901
- fix: prettynames qwen models were bogus by @starpit in #902
- feat(local): add model pool unload to release GPU memory by @starpit in #903
- refactor: remove prepare support by @starpit in #906
- feat(gce): add zone fallback for up() on ZONE_RESOURCE_POOL_EXHAUSTED by @starpit in #909
- chore: bump to leann-core v0.1.7 by @starpit in #908
- fix: update vllm-llmd 0.5.0 patch to latest IBM/vllm rebase by @starpit in #907
- chore(deps): update rust crate rustls to v0.23.37 by @renovate[bot] in #905
- chore: bump to 0.21.0 by @starpit in #910
- fix: update llm-d patch to pull in spnl 0.21.0 by @starpit in #911
Full Changelog: v0.20.1...v0.21.0
v0.20.1
What's Changed
- perf: reduce unnecessary heap allocations in core paths by @starpit in #886
- chore(ci): avoid running expensive musl test builds on PR (release only) by @starpit in #888
- feat(gce): add --force flag to vllm down to ignore not-found errors by @starpit in #887
- perf: reduce remaining heap allocations across spnl/src by @starpit in #889
- chore: bump to 0.20.1 by @starpit in #891
Full Changelog: v0.20.0...v0.20.1
v0.20.0
What's Changed
- fix(ci): wait for ollama server readiness after install by @starpit in #874
- feat(mistralrs): add local embedding support by @starpit in #875
- chore(deps): update rust crate pyo3 to v0.28.2 by @renovate[bot] in #873
- chore(bench): remove mt_rag benchmark by @starpit in #876
- fix(deps): update rust crate async-openai to 0.33.0 by @renovate[bot] in #871
- fix: for unused tokenizer field if k8s feature is not enabled by @starpit in #877
- feat(cli): add
spnl benchsubcommands by @starpit in #878 - feat(gce): shared defaults.env and precompiled wheel detection by @starpit in #879
- chore: update Homebrew formula to v0.19.0 by @github-actions[bot] in #880
- fix(ci): start ollama server manually when systemd is unavailable by @starpit in #881
- feat(bench): add multi-dimensional RAG evaluation metrics to ragcsv by @starpit in #882
- fix: another fix for ollama hang in core.yml workflow by @starpit in #884
- chore: bump to 0.20.0 by @starpit in #883
Full Changelog: v0.19.0...v0.20.0
v0.19.0
What's Changed
- feat(bench): add needle-in-a-haystack (NIAH) benchmark by @starpit in #857
- chore: update Homebrew formula to v0.18.1 by @github-actions[bot] in #858
- docs(benches): refactor benchmark documentation with comprehensive filtering guide by @starpit in #861
- chore(deps): update rust crate dirs to v6 by @renovate[bot] in #859
- ci: skip flash-attn CUDA builds on PRs by @starpit in #863
- feat(bench): add RULER benchmark - faithful Rust port of NVIDIA's long-context evaluation by @starpit in #864
- chore(deps): update rust crate clap to v4.5.59 by @renovate[bot] in #862
- ci: fix for buggy cuda filtering in release-cli.yml workflow by @starpit in #866
- ci: only use actions/cache for flash-attn cli builds by @starpit in #867
- feat(bench): add map-reduce chunking to NIAH benchmark by @starpit in #865
- ci: skip arm cli builds in pull requests by @starpit in #868
- chore: bump to llmd 0.5.0 by @starpit in #855
- fix(ollama): make retry loop actually retry all transient errors by @starpit in #872
- feat(bench): add ragcsv benchmark for CSV-based RAG evaluation by @starpit in #869
- chore: bump to 0.19.0 by @starpit in #870
Full Changelog: v0.18.1...v0.19.0
v0.18.1
v0.18.0
What's Changed
- chore(deps): update rust crate pyo3 to v0.28.1 by @renovate[bot] in #830
- chore(deps): update rust crate uuid to v1.21.0 by @renovate[bot] in #829
- ci: add Cargo build caching for CUDA release builds by @starpit in #836
- chore: remove useless CLAUDE.md file by @starpit in #835
- chore: update Homebrew formula to v0.17.0 by @github-actions[bot] in #833
- refactor(bench): consolidate benchmarks into spnl crate with Criterion by @starpit in #834
- chore: remove cli image support by @starpit in #840
- chore(deps): update rust crate criterion to 0.8 by @renovate[bot] in #839
- chore(deps): update actions/cache action to v5 by @renovate[bot] in #838
- chore(deps): update nvidia/cuda docker tag to v12.9.1 by @renovate[bot] in #831
- feat(vllm): add optional --tokenizer argument for k8s deployments by @starpit in #841
- fix: -F vllm without -F k8s or -F gce fails to build by @starpit in #848
- ci: publish_crate workflow incorrectly uses "publish_pypi" as name by @starpit in #847
- chore(deps): update rust crate futures to v0.3.32 by @renovate[bot] in #843
- chore(deps): update rust crate indicatif to v0.18.4 - autoclosed by @renovate[bot] in #837
- fix(openai): correct timing table display for multiple tasks by @starpit in #842
- fix: metal and cuda cli features should enable cli local feature by @starpit in #846
- chore: bump to 0.18.0 by @starpit in #852
Full Changelog: v0.17.0...v0.18.0
v0.17.0
What's Changed
- fix: change --time output to use stdout, and fix vllm-gce tests to use new --time feature by @starpit in #826
- chore(deps): update rust crate google-cloud-storage to v1.8.0 by @renovate[bot] in #825
- chore(deps): update rust crate google-cloud-compute-v1 to v2.2.0 by @renovate[bot] in #824
- chore(deps): update rust crate google-cloud-lro to v1.3.1 by @renovate[bot] in #823
- feat(local): add pretty model names and 'spnl list' command by @starpit in #827
- feat: add CUDA support via feature flags in cli package by @starpit in #822
- chore: bump to 0.17.0 by @starpit in #828
Full Changelog: v0.16.1...v0.17.0
v0.16.1
What's Changed
- chore: update Homebrew formula to v0.16.0 by @github-actions[bot] in #813
- feat: add TTFT and ITL metrics with tabled output by @starpit in #814
- feat: add SPNL_NUM_PARALLEL env var for bounded concurrency by @starpit in #815
- fix: local (mistralrs) backend should use default progress bar multiplier by @starpit in #812
- fix: improve table style in --time output by @starpit in #816
- docs: update README.md and docs/cli.md to document --time option by @starpit in #817
- refactor(mistralrs): use mistralrs::best_device() instead of custom device detection by @starpit in #819
- chore: bump to 0.16.1 by @starpit in #818
Full Changelog: v0.16.0...v0.16.1