Skip to main content

SLM Factory

SLM Factory is the packaging boundary for SLMCortex v0.1.

It turns an existing adapter directory plus provenance into a self-describing SLM package without retraining a model or mutating checked-in research artifacts.

Responsibilities

  • package an existing adapter into a deterministic SLM artifact
  • optionally run the product train-slm wrapper for built-in research SLMs
  • record provenance, checksums, protected input snapshots, and composition metadata
  • validate emitted package files for internal consistency

Inputs

  • adapter directory with adapters.safetensors or adapter.gguf
  • train and eval dataset paths for provenance
  • eval summary JSON
  • optional examples and composition metadata

Outputs

One package directory containing slm.yaml, metadata.json, training_config.json, eval.json, README.md, and adapter/.

SLM Factory owns the package-slm and product train-slm stages of the demo flow.