Skip to content
ignisignis

ignis v0.9.2

Released 2026-10-29. The container image, the archives and the notes as written: the release on GitHub.

The v0.9.0 and v0.9.1 downloads were not GPU builds: use this one. Their Linux tarball, Windows zip and container images were built without the cuda feature. The server started, reported ready and answered with the CPU mock’s placeholder text whatever the model, logging only a WARN ignis.model.mock_compute (“built without –features cuda”). v0.9.2 is the same code as v0.9.1, built with CUDA again. If you installed 0.9.0 or 0.9.1, replace the binaries or the image with these. Your downloaded models and your config stay as they are. (#408)

Changes

Section titled “Changes”
  • Release builds are CUDA builds again (#408).
    • Cause: cargo read --features ignis-server/cuda as a feature of the server’s dev-dependency on itself, which a build of the binaries never compiles, so the server came out without CUDA and nothing failed.
    • Fix: the builds now pass --features cuda.
    • New guard: the release pipeline refuses to package a server that would fall back to the CPU mock (mk/cuda-build-check.sh, with its own selftest), and a test holds every build recipe to the working spelling.

Container image (linux/amd64)

Section titled “Container image (linux/amd64)”
podman run --rm --device nvidia.com/gpu=all -p 8000:8000 \
  -v /path/to/models:/models:ro \
  -e IGNIS_MODEL_ARTIFACT=/models/<name>.ninfer \
  ghcr.io/gpillon/ignis:0.9.2

(docker: --gpus all in place of --device.) The image carries the CUDA runtime but no driver — the host’s NVIDIA driver is injected by the container runtime. It is built from the same compile as the Linux tarball below, so the two always carry the same binaries.

Binaries

Section titled “Binaries”

The kernel is compiled for SM120a (RTX 5090 / Blackwell consumer) only, and the engine needs a .ninfer artifact, which is not part of this release. The Linux tarball needs the CUDA 13 runtime on the host; the Windows zip carries cudart64_*.dll. Both need an NVIDIA driver new enough for CUDA 13.

Apache-2.0 (LICENSE, NOTICE); the vendored reference ops keep their own terms (NOTICE-kernel).

Full Changelog: https://github.com/gpillon/ignis/compare/v0.9.1...v0.9.2