Qdrant tests Perplexity's pplx-embed-v2-late: 21x smaller vectors with 99.9% retrieval quality retained
qdrant_engine · x · 2026-10-14
Qdrant benchmarked Perplexity's pplx-embed-v2-late embeddings with its TurboQuant compression across 4 ViDoRe v3 datasets: vectors shrink from 856 KB to 40 KB per page (21x smaller), retaining 99.9% of baseline retrieval quality with rescoring. The trade-off: rescoring requires keeping the original vectors. Full benchmarks in the thread.
Related event: Qdrant Shrinks Perplexity's New Embedding Model 21x With TurboQuant(2 posts)→
More from Infra
- LightVAE speeds up video VAE decoding up to 2.24x on H100 with minimal quality loss — Shiqiao Gu · 2026-10-14
- Orbital data centers have no off-switch: Reddit debates AI safety paradox — Zappy_Oh · 2026-10-14
- Rednote's Dots unveils MSAR: 1.58x iso-loss compute speedup over residual baseline — teortaxesTex · 2026-10-14
- VirusTotal ships macOS Mirage sandbox to analyze fake password-prompt infostealers on Apple Silicon — jedisct1 · 2026-10-14
- ClickHouse merges IPCrypt support for GDPR-friendly IP pseudonymization — jedisct1 · 2026-10-14
- 4-bit keeps 98-99% quality, DeepSeek V4 Flash isn't a local-deployment model — gnukeith · 2026-10-14