Mohamed Arbi Nsibi·Oct 31Qdrant 1.20: BM25 Without Sparse Vectors, Turbo16 & Turbo8BM25 straight from your payload, a tokenizer that finally understands camelCase, nested filters 7x faster, and two new TurboQuant…
InData Science CollectivebyManjunath Janardhan·Sep 25I Measured Prism ML’s Bonsai 2 Against My Own Builds of Alibaba’s Qwen3.8–27B.36 tokens per second in about 10 GB on an M4 Max, the same test scores as a 4-bit build twice its size, and five traps that nearly skewed…A response icon2A response icon2
InData Science CollectivebyManjunath Janardhan·Sep 12Muse Glimmer on a 16 GB Mac mini: the 5 GB your 4-bit build never compressedMeta’s Muse Glimmer compresses to 14.88 GiB at four bits. The standard 4-bit build of the same model weighs 19.88 GiB. Chasing that…
Mohamed Arbi Nsibi·Sep 4Qdrant 1.19: Turbo4 Datatype & Memory TiersOne config line, nine times less disk, and a single knob for all your RAM decisions. Here’s what’s actually new and when you should care.
DhanushKumar·Aug 20TurboVec and TurboQuant: How 10 Million Vectors Can Fit in a Few GigabytesVector search has a problem that most RAG tutorials tend to hide.
MELVIN MARSHAL PEREIRA·Aug 19TurboQuant Compression For LLM KV Cache and Enterprise AI: A Comprehensive AnalysisFrom Memory Compression to Practical Reality
InData Science CollectivebyManjunath Janardhan·Aug 17Qwen 3.8–27B on a 16 GB Mac mini: Alibaba’s new vision model, fully in memoryQwen ships it as a 55.6 GB download and claims it beats Claude Opus 4.6 on agentic benchmarks. Compressed to 11.55 GB — vision tower…A response icon9A response icon9
ThamizhElango Natarajan·Aug 16TurboQuant, Explained: How Google’s KV Cache Compression Algorithm Actually WorksIntroduction
InMac O’ClockbyManjunath Janardhan·Jul 25Run Laguna S-2.1 Locally on a Mac: a 118B Coding Model in 27 GB, Fully in MemoryPoolside’s Laguna S-2.1 is a 118-billion-parameter model built for agentic coding. Compressed to 27 GB with 1.58-bit quantization, it runs…A response icon4A response icon4
Adeel Tahir·Jul 24ForgeVector: A Vector Database on Top of TurboQuantA while back, I came across a post about Google’s TurboQuant.
