Benchmark Runs¶
See docs/maintainer/benchmark_truth_table.md for what each benchmark measures and does not claim.
Baselines¶
- Baselines live under
benchmarks/baselines/. - The CI matrix expects entries for memory, FAISS, and Qdrant (exact + ann).
- Replace
status: todoplaceholders after running on a dedicated host.
Run Locally¶
make install
.venv/bin/python -m bijux_vex.boundaries.cli.app bench --store vdb --vector-store faiss --mode exact
.venv/bin/python -m bijux_vex.boundaries.cli.app bench --store vdb --vector-store faiss --mode ann
Qdrant (Local)¶
Qdrant benchmarks can run without Docker using the in-memory client:
.venv/bin/python -m bijux_vex.boundaries.cli.app bench --store vdb --vector-store qdrant --vector-store-uri :memory: --mode exact
.venv/bin/python -m bijux_vex.boundaries.cli.app bench --store vdb --vector-store qdrant --vector-store-uri :memory: --mode ann
Qdrant (Docker)¶
If you prefer a local service:
docker run --rm -p 6333:6333 qdrant/qdrant:latest
Then run the same commands without --vector-store-uri :memory:.
TODOs¶
- TODO: generate FAISS exact/ann baselines on a faster host.
- TODO: generate Qdrant exact/ann baselines on a faster host.
- TODO: generate FAISS 100k baselines on a faster host.