CiteMesh corpus indexes

Portable retrieval-document embeddings with normalized paper metadata for CiteMesh. See the repository for installation and usage. No model weights are included. Query and graph encoding use the selected model.

embeddinggemma-512-int8

3,224,542 active papers; 512 dimensions; int8 storage. Snapshot: 21a34a9d-37fc-4832-a676-f303e8db2583. Source: librarian-bots/arxiv-metadata-snapshot, revision c428226b689a0e17db2cd55265133b9eca15d97f, split train. See the source dataset for attribution and licensing. Model: unsloth/embeddinggemma-300m, revision bfa3c846ac738e62aa61806ef9112d34acb1dc5a. Document formatter: b821375defea16a6; normalization: l2-v1; quantization: affine-int8-v1.

Install a configuration

citemesh cache import --bucket <this-bucket> --config <configuration>
citemesh cache scan

Default configuration: embeddinggemma-512-int8. Each configuration retains its current snapshot; older snapshot directories are removed after publication.

Use the installed namespace with citemesh search --mode local --cache-namespace <namespace> "your query". Run citemesh cache refresh --namespace <namespace> to reconcile newer source data locally.