69 lines (48 loc) · 2.45 KB

Scripts

This page maps each recommended script to its role.

Recommended Entry Scripts

Script	Use it for	Notes
`scripts/flat_lme_build_index.sh`	Build LongMemEval flat expansion caches	Generates summary, keyphrase, and userfact caches
`scripts/graph_lme_construct.sh`	Build LongMemEval graph artifacts	Recommended graph indexing entry
`scripts/graph_lme_run_retrieval.sh`	Run LongMemEval graph retrieval	Clean named-argument interface
`scripts/halu_run.sh`	Run HaluMem flat memory evaluation	Main flat HaluMem entry; final QA-eval scoring is a separate step
`scripts/graph_halu_construct.sh`	Build HaluMem graph artifacts	Graph indexing for HaluMem
`scripts/graph_halu_run_retrieval.sh`	Run HaluMem graph retrieval	Produces retrieval JSON
`scripts/graph_halu_eval_pipeline.sh`	Run end-to-end HaluMem graph evaluation	Best single-command graph pipeline

LongMemEval

Flat

scripts/flat_lme_build_index.sh
scripts/lme_run_retrieval.sh

Recommended pattern:

use flat_lme_build_index.sh for caches
use python -m src.flat.lme_run_retrieval for explicit retrieval arguments

Graph

scripts/graph_lme_construct.sh
scripts/graph_lme_run_retrieval.sh

HaluMem

Flat

scripts/halu_run.sh

Graph

scripts/graph_halu_construct.sh
scripts/graph_halu_run_retrieval.sh
scripts/graph_halu_eval_pipeline.sh

Lower-Level Python Entry Points

Use these when you want named arguments or to compose your own workflow:

Module	Purpose
`python -m src.flat.lme_run_retrieval`	Flat LongMemEval retrieval
`python -m src.graph.lme_construct_graph`	LongMemEval graph construction
`python -m src.graph.lme_run_retrieval`	LongMemEval graph retrieval
`python -m src.flat.halu_run`	Flat HaluMem evaluation
`python -m src.graph.halu_construct_graph`	HaluMem graph construction
`python -m src.graph.halu_run_retrieval`	HaluMem graph retrieval
`python -m evals.run_generation`	Generation for retrieval outputs
`python -m evals.lme_compute_qa`	QA evaluation for LongMemEval outputs
`python -m evals.halu_graph_eval`	HaluMem graph intermediate generation
`python -m evals.halu_eval`	HaluMem final scoring

Documentation Rule for This Repo

When in doubt:

prefer scripts for the common path
prefer Python modules for explicit named parameters
prefer docs/ over expanding the root README again