Last 12 weeks · 8 commits
5 of 6 standards met
Summary The value of the "sec" parameter is a float of seconds not microseconds. Changed to "seconds" so relativedelta is computed correctly. This was causing any time longer than 1 second to be formatted as "0 seconds". See docs here https://dateutil.readthedocs.io/en/stable/relativedelta.html Test Plan I have currently not added any test coverage for this issue
Summary Stop storing PDQ hex strings in ; keep only match metadata (typically ), since FAISS already holds the vectors. Migrate legacy pickled sidecars on unpickle via so existing HMA index blobs keep working until rebuild. Add coverage that new pickles store metadata-only and that legacy sidecars still query correctly. This should lower steady-state matcher RSS for large PDQ indexes (one fewer Python string per hash in the sidecar). It does not address FAISS refresh/leak behavior. Test plan [x] [x] and [ ] Confirm HMA matcher can load an existing pickled index LOB and return matches [ ] After indexer rebuild, confirm new LOBs pickle metadata-only sidecars and memory footprint drops vs baseline
Repository: facebook/ThreatExchange. Description: Trust & Safety tools for working together to fight digital harms. Stars: 1373, Forks: 354. Primary language: C++. Languages: C++ (73.8%), Python (16.7%), Java (3.8%), JavaScript (1%), Ruby (0.9%). Homepage: https://facebook.github.io/ThreatExchange/ Topics: hashing, image, image-hashing, image-similarity, ncmec, perceptual-hashing, stopncii, threatexchange, video, video-hashing, video-similarity. Latest release: hma-v1.1.7 (2mo ago). Open PRs: 6, open issues: 45. Last activity: 2d ago. Community health: 87%. Top contributors: Dcallies, johnkerl, BarrettOlson, dmukhg, mgoffin, jeberl, jessek, juanmrad, dependabot[bot], mengyangwang and others.