Skip to main content

No-Model Demo

The no-model demo is the safest first proof that SLMCortex is wired correctly.

It validates the product flow without downloading model weights or running real inference.

Run It​

DEMO_ROOT="$(mktemp -d "${TMPDIR:-/tmp}/slmcortex-demo.XXXXXX")"
python scripts/run_slmcortex_demo.py --output-root "$DEMO_ROOT"

What It Checks​

  • packaging existing adapters into self-describing SLM packages
  • composing those packages into one runtime bundle
  • validating the runtime bundle before execution
  • dry-run routing for inference
  • bounded agent control flow against a local repository

Expected outputs under $DEMO_ROOT:

python_slm/
debugging_slm/
runtime/
agent-trace.json

Why This Matters​

This is the shortest route to answering the practical question:

"Does the product flow work on this machine at all?"

If the demo fails, fix that first. It is much cheaper than debugging real model downloads, backend installs, or local training issues at the same time.

If you want a package-first smoke for an arbitrary SLM ID:

python scripts/run_slmcortex_arbitrary_slm_smoke.py

If you want the slower real-training variant:

python scripts/run_slmcortex_arbitrary_slm_smoke.py --real-training

Treat those as follow-up validation, not the default first step.