micm_nlp.evals

Evaluation: metric assembly, prediction post-processing and plots.

• eval — get_compute_metrics builds the compute_metrics callable the Trainer expects; get_preprocess_logits_for_metrics is its companion hook, which runs before logits are accumulated. Also handles per-task grouping and label restriction.

• plot — confusion matrices.

• metrics — the metric implementations evaluate does not provide directly. See that package’s docstring: it holds two different module shapes, on purpose.

Submodules

• eval
• plot
• metrics