Skip to content
AI-LinkMO

Install notes: compilers, CUDA, corporate indexes

The Install step (uv sync --extra test) is enough on most machines. These notes cover the cases where it is not.

Red Hat hosts with an older gcc

Section titled “Red Hat hosts with an older gcc”

Some of the compiled dependencies need gcc 12 or later. On RHEL-family hosts:

Terminal window
sudo yum install gcc-toolset-12 gcc-toolset-14 ninja-build -y   # once, if gcc < 12
scl enable gcc-toolset-12 bash

Builds behind a corporate package index, or with CUDA

Section titled “Builds behind a corporate package index, or with CUDA”
Terminal window
# export UV_DEFAULT_INDEX=https://proxy.example.com/repository/pypi-all/simple
uv cache clean
MAX_JOBS=4 UV_HTTP_TIMEOUT=60s TORCH_CUDA_ARCH_LIST="8.6" uv sync --extra test   # be patient

MAX_JOBS limits parallel compilation. TORCH_CUDA_ARCH_LIST should match your GPU. Neither is needed for a CPU-only install from the public index.

Regenerating the web UI’s schema

Section titled “Regenerating the web UI’s schema”

The web UI includes a JSON Schema derived from the ontology. If the ai-atlas-nexus package has been upgraded (or you are unsure), rebuild the schema. The path below works for any version:

Terminal window
SCHEMA="$(uv run python -c 'import ai_atlas_nexus, pathlib; print(pathlib.Path(ai_atlas_nexus.__file__).parent / "ai_risk_ontology" / "schema" / "ai-risk-ontology.yaml")')"
uv run gen-json-schema --stacktrace --preserve-names --mergeimports "$SCHEMA" > lib/frontend/static/schema/ai-risk-ontology.json

The two logos

Section titled “The two logos”

lib/frontend/static/mylogo.png is the logo the web UI serves. Replace it to rebrand the UI. .assets/mylogo.png is the example logo for a bring-your-own-data identity, and the UI does not read it.