A volume or bind mount from the 0.1.x image keeps its models in /data/models. The entry point moves a folder that is part of the volume, but a folder the person
mounted there is left alone, and a failed move leaves the files too; with no MODEL_PATH the start then did not see them and downloaded the model again
(tens of GB) without a word. Now, when .gguf files are still in /data/models and MODEL_PATH is not set, the container stops with code 2 and names the folder,
says what to do (MODEL_PATH to the folder with the shards, or the files into /data/Strata-data/models) and that nothing was downloaded.
ALLOW_DOWNLOAD=1 goes on anyway with the same text as a warning. docs/DOCKER.md lists the setting. The other points of the issue (the key, the removed
variables, the status loop) are the owner's design questions and are not changed here. Tested by running the guard function with sh (tests/terminal/test_docker_files.py);
the whole container was not run (no docker on this PC).
docker/build.sh staged only top-level tools/*.py, so the Terminal stage stopped at 'installer tool missing: tools/dsv41/tune.py'.
Ported from PR #3974 by hurui200320: the pathspec is tools/**/*.py, the no-.git fallback finds nested files, and the context test reads
INSTALLER_TOOLS and INSTALLER_SCRIPTS from packaging/package_terminal.py. Fixes the failure that images built from v0.2.4.1 and v0.2.4.2 had.
Cargo.toml and Cargo.lock, desktop and host package.json and lock, PRODUCT_VERSION, CMake project and Metal STRATA_VERSION, build_hip_linux PKGVER default, About comments, tests that pin the number, Docker image tags and docs (0.2.3-<flavor>), download names in the READMEs, AMD_HIP.md and UPDATES.md, AGENTS.md, the bug report placeholder. The engine moves too: it shares the CMake number and its source changed since 0.2.2 (DFlash2, verify and prefill paths). A new test checks every place agrees and that a 0.2.2 install (Control 0.1.47) is offered Control 0.1.48.
Cargo.toml and Cargo.lock, desktop and host package.json, PRODUCT_VERSION, CMake project and Metal STRATA_VERSION, build_hip_linux PKGVER default, About comment, tests that pin the number, Docker image tags and docs, download names in the READMEs, AMD_HIP.md and UPDATES.md.
package_terminal.py reads STRATA_BUILD_COMMIT first and only then asks git, and no longer crashes when git is missing. build.sh passes the commit as a build arg, the Dockerfile turns it into that env.
- The EXIT trap of build.sh (--ctx-only or --ctx given) and publish.sh (--work given) ended with a failing test, so dash reported exit 1 after a
successful run. Both traps now use if/fi. Test: build.sh --ctx-only exits 0 and its context holds docs/TERMINAL_LAUNCH.md.
- docs/DOCKER.md: image sizes, the 2026-10-19 test of the images built from the 0.2.0-beta packages (cuda13 and cuda12 on an RTX A4000, ROCm images
built and checked, wrong-card message).
- .github/workflows/docker-images.yml: manual only (workflow_dispatch) with the release tag as input; a plan job lists the flavors the
release has an engine package for; one job per flavor builds the image and, only with push on, pushes ghcr.io/niko1221/strata:<version>-<flavor>
and latest-<flavor> with GITHUB_TOKEN (packages: write).
- docker/fetch-engines.py: downloads the engine package from the release and checks it against the signed update feed (feed signature, archive
size and SHA-256, manifest hash, package signature, every file), the same code as `strata update`. Flavors cuda13, cuda12, rocm-gfx120x,
rocm-gfx110x, rocm-gfx1151 (the halo package), sycl (only if the feed lists one).
- docker/publish.sh: the same steps by hand (fetch, build, start check, push only with --push).
- Fix: docker/build.sh did not copy docs/TERMINAL_LAUNCH.md, which package_terminal.py needs; no image could be built from a clean tree.
- Dockerfile: OCI source/url/documentation labels (ghcr links the package to the repository).
- Compose files and docs/DOCKER.md name the 0.2.1 tags on ghcr.io; DOCKER.md has a maintainers section. Tests for the scripts, the workflow and the fetch.
Niko1221