v0.34.0

0.34.0 (2026-10-08)

Features

• models: retire superseded MLX candidates (589ef48)
• models: retire superseded MLX candidates (820c500)

Bug Fixes

• docs: fit model currency within CI limits (451ccc6)

v0.33.3

0.33.3 (2026-10-07)

Bug Fixes

• authenticate model preflight requests (64c2548)
• authenticate model preflight requests (1a0482c)

v0.33.2

0.33.2 (2026-10-07)

Documentation

• update NVIDIA benchmark guidance (#301) (66a24a1)

v0.33.1

0.33.1 (2026-10-07)

Bug Fixes

• benchmarks: require enforced GPU power limit at publish (#299) (47a324a)

v0.33.0

0.33.0 (2026-10-07)

Features

• benchmarks: carry model task metadata (56469b7)
• schema: add Stage 0 Mac load phase contract (a03a5cf)
• schema: add Stage 0 Mac load phase contract (6182e10)

Bug Fixes

• benchmarks: reconcile model-task metadata schema (1d6ef02)
• docs: keep schema guide within size limit (ea66812)
• publish: require Stage 0 catalog task alignment (30917f4)

Documentation

• schema: split Stage 0 load phase guidance (4e6f8d2)

v0.32.0

0.32.0 (2026-10-07)

Features

• publish pinned Hub metadata with benchmark results (#295) (46628f1)

v0.31.0

0.31.0 (2026-10-07)

Features

• converters: add llamacpp-server request-log converter (#292) (5a6d897)

v0.30.1

0.30.1 (2026-10-06)

Bug Fixes

• space: follow the dark theme for the page and status panel (#290) (9d13692)

v0.30.0

0.30.0 (2026-10-06)

Features

• configs: declare GPQA-Diamond and LiveCodeBench campaign recipes (#288) (a1ad810)
• converters: keep campaign dimensions on every envelope (#287) (e165251)

Bug Fixes

• dataset: serve the default config from a rolling recent file (#285) (4a66793)

v0.29.5

0.29.5 (2026-10-05)

Bug Fixes

• agentic: measure streamed output timings (ff8c451)
• ci: wire shared gate inputs (565c46f)