sveneziale/activations-and-barcodes

sveneziale/activations-and-barcodes

Compute artifacts pushed by tda-for-llms's Hugging-Face-backed pipeline (hf.enabled: true in experiment.yaml).

Layout

Two top-level folders:

activations/{model_slug}/{corpus}/{revision}/{act_name}/

Raw per-cloud activation matrices extracted from the model, one independent copy per checkpoint (revision). Independent of `topology.metric` — the same activations are reused across every metric or topology config that shares the same model + corpus + checkpoint + activation stream, so they are only ever GPU-extracted once.

• —activation_metadata.csv — one row per cloud (cloud_id, revision, layer, head, n_prompts, ...).
• —activations.npz — raw matrices, keyed act_{cloud_id} matching the cloud_id column above.

barcodes/{model_slug}/{corpus}/{revision}/{act_name}/{persistence|zigzag}/{metric}/

Persistent-homology outputs computed from the activations above. One copy per topology metric/mode, since the barcode depends on it.

• —persistence_summaries.csv — one row per cloud, scalar topological feature summary.
• —persistence_diagrams.npz — raw barcodes, keyed cloud_{cloud_id}_h{dim}.
• —manifest.json — full run provenance (model, revisions, activation stream, corpus, topology parameters).