From 7048d0cd54b652fd126fdd4f70cb6112a15220c2 Mon Sep 17 00:00:00 2001 From: Abdon Pijpelink Date: Fri, 8 May 2026 08:25:09 +0200 Subject: [PATCH] Remove links to article --- qdrant-landing/content/blog/qdrant-1.18.x.md | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/qdrant-landing/content/blog/qdrant-1.18.x.md b/qdrant-landing/content/blog/qdrant-1.18.x.md index 9c23deb1c..ae05afb73 100644 --- a/qdrant-landing/content/blog/qdrant-1.18.x.md +++ b/qdrant-landing/content/blog/qdrant-1.18.x.md @@ -38,8 +38,6 @@ Version 1.18 introduces support for [TurboQuant](/documentation/manage-data/quan Qdrant's implementation of TurboQuant extends the original algorithm to close the gap between the algorithm's theoretical assumptions and real-world embeddings. A length renormalization step corrects a recall-degrading bias caused by quantization error, an idea borrowed from [RaBitQ](https://arxiv.org/abs/2405.12497). A per-coordinate calibration pre-pass fits the data to precomputed codebooks, aiming to recover accuracy lost to distribution mismatch. Cosine, dot product, and L2 are all supported as first-class distance metrics. And finally, we implemented highly optimized SIMD acceleration for TurboQuant to achieve maximum performance. -To learn more about Qdrant's TurboQuant implementation, refer to [our article](/articles/turboquant-quantization/). - ### How TurboQuant Compares #### TurboQuant vs Scalar Quantization @@ -68,7 +66,7 @@ The following table shows recall@10 for 1-bit TurboQuant (TQ1) compared to uncom Compared to 1-bit binary quantization, 1-bit TurboQuant offers better recall at equivalent storage budgets, albeit at a lower speed. Similar trends are observed for 1.5-bit and 2-bit configurations. -Detailed numbers including throughput and indexing times are [in our article](/articles/turboquant-quantization/). +We will soon publish an article with detailed numbers, including throughput and indexing times. ### Get Started with TurboQuant