O
Ollama
AI-ASSISTANTS
Velocity2.2
Run and manage large language models locally on your own machine.
Heads-down in a long v0.30.0 rc cycle, hardening Windows and ROCm builds
local-llmrelease-candidateswindows-buildsrocmrunner-refactor
◆Current state
Ollama is deep in a v0.30.0 release-candidate run (rc20 through rc27), and the visible churn is almost entirely platform plumbing: Windows CPU build workarounds, ROCm CI cache fixes, and clearer Windows exit logging. The one structural signal is an upstream merge into a 'llama-runner-phase-0' branch, hinting at runner work underneath the build noise.
◆Where it's heading
The sustained rc cadence plus the 'llama-runner-phase-0' branch name point to an in-progress rework of the model runner, with cross-platform build reliability (Windows CPU, ROCm) as the gating concern before a stable cut. Nothing user-facing has landed in these entries; this is stabilization, not feature work.
◆Prediction
Expect a stable v0.30.0 once the rc series settles, likely carrying the reworked llama runner. Near-term entries will stay build- and CI-focused.
◆Recent moves
- 1d ago
Windows CPU build path workaround in CI
A CI-only path workaround to unblock the Windows CPU build during the v0.30.0 rc series. Pure build plumbing with no user-visible effect.
View source ↗ - 1d ago
Merge upstream into llama-runner-phase-0 branch
An internal merge of upstream main into the llama-runner-phase-0 branch, resolving conflicts in server image handling. Notable only as a marker that the runner-refactor branch is being kept in sync.
View source ↗ - 8d ago
Clearer Windows exit error logging
Minor diagnostic polish that improves error logging on Windows shutdown. Helps debugging but changes nothing users interact with.
View source ↗ - 9d ago
Fix ROCm build cache miss in CI
A CI cache fix for the ROCm build path that speeds builds but is invisible to users. Fits the stabilization theme of the rc run.
View source ↗