Qdrant tests Perplexity's pplx-embed-v2-late: 21x smaller vectors, 99.9% retrieval quality kept
qdrant_engine · x · 2026-10-14
Qdrant tested Perplexity's new late interaction retriever pplx-embed-v2-late with Qdrant's TurboQuant across 4 ViDoRe v3 datasets.
- Vectors shrink 21x: 856 KB → 40 KB per page
- 99.9% of baseline retrieval quality retained with rescoring
- Trade-off: rescoring requires keeping the original vectors
A notable storage-cost optimization for large-scale retrieval/RAG setups; full benchmarks in the original thread.
Related event: Qdrant Tests Perplexity's Late-Interaction Embeddings: 21x Smaller Vectors(3 posts)→
More from Infra
- Investor warns undifferentiated neoclouds pivoting to inference face consolidation — AccBalanced · 2026-10-14
- Lightning AI Joins GTC Berlin Panel on GPU Access for AI Startups at Every Stage — LightningAI · 2026-10-14
- Hugging Face to livestream how speculative decoding delivers multipliers in local LLM speed-ups — mervenoyann · 2026-10-14
- The Jensen Huang Playbook: How NVIDIA Built CoreWeave as a Buffer Against Its Biggest Customers — AccBalanced · 2026-10-14
- 64GB Nvidia Spark spotted at €6,574 in Europe — €900 more buys the 128GB model — AIFlow_ML · 2026-10-14
- Toby Ord says RL post-training compute neared parity with pretraining; bet on broad compute-capability trends — ben_j_todd · 2026-10-14