naeural_core/
houses the runtime:
main/
orchestrates startup,
business/
contains pipeline plugins and the testing harness,
comm/
handles transports, and
serving/
wraps model execution. Runtime artifacts land in
_local_cache/
, created on demand. Optional packages live under
extensions/
, docs under
docs/
, and hardware spikes stay in
xperimental/
. Use
start_nen.py
as the launcher when embedding the core into Edge Node deployments.
python -m venv .venv && source .venv/bin/activate
: provision a Python 3.10–3.11 virtual environment.pip install -e .
followed by pip install -r requirements.txt
: install the package in editable mode plus heavyweight plugin dependencies.python start_nen.py
: boot the orchestrator; expects config_startup.json
or an EE_CONFIG
JSON plus .env
secrets loaded via ratio1.utils.load_dotenv
.python naeural_core/serving/model_testing/test_all/test_all_servings.py
: run the GPU-backed serving smoke tests; set EE_MINIO_*
credentials and ensure CUDA availability.python xperimental/minio/test_minio.py
(pattern): execute targeted experiments without affecting release modules; document prerequisites alongside the script.
Match the repository’s two-space indentation,
snake_case
functions,
PascalCase
classes, and all-caps constants (imported as
naeural_core.constants as ct
). Prefer double quotes for user-facing strings. Reuse the shared
Logger
for structured logs and reference config keys through
ct
instead of raw literals. When exposing HTTP endpoints, wrap plugin methods with
@BasePlugin.endpoint
and keep companion templates in the same package (see
docs/FastApiWebApp.md
).
Keep fast checks close to their modules—
naeural_core/business/test_framework/
is the staging ground for unit-style tests. Name files
test_<feature>.py
so
python -m unittest discover
remains viable. Integration suites under
naeural_core/serving/model_testing/
rely on CUDA hardware, MinIO access, and populated
.env
files; record the environment, datasets, and command output in the PR. Experimental proofs should stay in
xperimental/
with a module-level docstring describing required GPUs or services.
Follow the Conventional Commit voice observed in history (
feat:
,
fix:
,
chore:
). Reference issues with
(#123)
when applicable, cap subjects at 72 characters, and detail behavioral impact plus commands executed in the description. Attach logs or screenshots for inference-facing changes, and call out any new ports, environment keys, or migrations. Verify that
.env
files and cached artifacts stay untracked.
Runtime configuration resolves from
config_startup.json
,
EE_CONFIG
, and
EE_ID
; keep secrets outside the repo and feed them through
.env
. Validate external payloads using the filters defined in
naeural_core.constants
when extending comms modules, and avoid embedding credential defaults inside plugin code.