Repository navigation
v0.9.11
·
1695 commits
to master
since this release
Provider v0.9.11 (Swift CLI)
Source: 2b714c4
Build: https://github.com/Layr-Labs/d-inference/actions/runs/36371161462
Binary SHA-256: 5c6e661290bd2a9e458348b53cc64025c0e51ba04cb8f5e3df9da47b1217452a
CodeDirectory SHA-256: a4a9440e33c940c86fded8511ebebd2608114bb1a44e9307946ddc7280fc6eb6
Bundle SHA-256: d3e35441dfbefbbf91bd4c3c1cb2465babffb562fee8651edfbc62f529f1fdf7
Metallib SHA-256: 38ceb8a1113b373ccaa89355d895fc8ded3bafc81076dc4ffdb426e2eee03a93
Signed and notarized: yes
Minimum macOS: 14.0
Install
curl -fsSL https://api.darkbloom.dev/install.sh | bashWhat's Changed
- Prefix cache: fix the hit-rate collapse (epoch churn, demand-gated writes, 1,024-token checkpoints, credit band, bounded fence, index sizing, sidecar parity) by @Gajesh2007 in #1207
- chore: remove pre-0.9.5 compat code, dead tests and Python/Rust/Solana-era leftovers by @Gajesh2007 in #1208
- feat(prefix-cache): chunk-agnostic recurrent capture and fork-target retention for Qwen donors by @Gajesh2007 in #1221
- feat(provider): enable SSD prefix caching and paged KV by default for Qwen3.5-9B by @Gajesh2007 in #1223
Full Changelog: v0.9.10...v0.9.11