CLI Command Group

moro eval

Execute deterministic schema gates, semantic LLM accuracy evaluations, Wasserstein representation drift audits, and adversarial stress testing.

Subcommands

Executes a structured evaluation test suite against an adapted model checkpoint or base foundation model.

moro eval run --checkpoint ./runs/run_01/checkpoints/best/ --suite ./eval/suite.yaml

Flag	Type	Default	Description
--checkpoint	Path	Required	Path to candidate model weights or adapter directory.
--suite	Path	./eval/suite.yaml	Path to evaluation suite YAML file.
--judge-model	String	local	Model used for semantic scoring: `local` (Ollama) or HF model ID.

Runs identical benchmarks across a base model and adapted checkpoint to measure accuracy delta and drift:

moro eval compare \
    --base Qwen/Qwen2.5-1.5B \
    --candidate ./runs/run_01/checkpoints/best/ \
    --suite ./eval/suite.yaml