18 comments

  • lmeyerov11 minutes ago
    Interestingly, while we don&#x27;t fine-tune generative models for Louie.ai, we found fine-tuning embedding models to be a major $ saver. Instead of 1K-2K wide frontier embedding vector lens... Just 64. Huge savings on vector DB $$$.<p>I&#x27;m curious how that works with something like turboquant. Not needed any more, still dominant, better together, ... .
  • Eridrus6 hours ago
    FAISS is no longer close to SoTA:<p><a href="https:&#x2F;&#x2F;ann-benchmarks.com&#x2F;index.html" rel="nofollow">https:&#x2F;&#x2F;ann-benchmarks.com&#x2F;index.html</a> <a href="https:&#x2F;&#x2F;vector-index-bench.github.io&#x2F;" rel="nofollow">https:&#x2F;&#x2F;vector-index-bench.github.io&#x2F;</a> <a href="https:&#x2F;&#x2F;big-ann-benchmarks.com&#x2F;neurips23.html" rel="nofollow">https:&#x2F;&#x2F;big-ann-benchmarks.com&#x2F;neurips23.html</a>
    • nl3 hours ago
      I think their point is the size&#x2F;performance tradeoff rather than outright performance. The point of TurboQuant is the size savings, while still giving high accuracy.<p>It&#x27;s been a while, but I do recall some high-performing vector matching indexes being very large.
    • ehsanu157 minutes ago
      Surprised that usearch isn&#x27;t in any of these, it&#x27;s pretty fast.
  • ghm21998 hours ago
    Wow! 4GB for 10 million documents. This means one could build a reverse index much faster than before and devx processes like debugging, performance testing would become much smoother. Can&#x27;t wait for the sqlite bindings to come out!
    • ghm21998 hours ago
      Also the removal latency is on a log scale. Which is quite insane.
  • nharada8 hours ago
    It would be nice to have the README be a little more human written for a project where you actually want people to adopt it
    • badatnames8 hours ago
      Anthropic employee. This is what your brain on kool aid looks like
      • deeviant7 hours ago
        Then again, if the only thing the human doing is bitching about AI use, it&#x27;s not really that comparatively useful.
        • righthand3 hours ago
          Sure it is useful, the bitching is canary in the shit software mine. How do you know the software isnt shit if the Readme is shit?
  • bobmarleybiceps6 hours ago
    people should read turboquant&#x27;s open review comments: <a href="https:&#x2F;&#x2F;openreview.net&#x2F;forum?id=tO3ASKZlok" rel="nofollow">https:&#x2F;&#x2F;openreview.net&#x2F;forum?id=tO3ASKZlok</a>
    • esafak3 hours ago
      tl,dr: there is an allegedly better alternative, and it&#x27;s already implemented everywhere: <a href="https:&#x2F;&#x2F;github.com&#x2F;VectorDB-NTU&#x2F;RaBitQ-Library#rabitq-in-industry" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;VectorDB-NTU&#x2F;RaBitQ-Library#rabitq-in-ind...</a>
  • anishvarghese8 hours ago
    This looks perfect for local, privacy first search, but since it&#x27;s built in Rust, has anyone tried compiling it to WASM to run directly inside a browser extension?
    • westurner7 hours ago
      oxirs does embeddings and GraphRAG, and full text search with Tantivy; oxirs-vec, oxirs-graphrag<p>There&#x27;s an oxirs-wasm with RDF and SPARQL bindings with a query budget. Tantivy-wasm says that the release WASM bundle is 1.5 MB.<p>cool-japan&#x2F;oxirs: <a href="https:&#x2F;&#x2F;github.com&#x2F;cool-japan&#x2F;oxirs" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;cool-japan&#x2F;oxirs</a><p>oxirs-wasm: <a href="https:&#x2F;&#x2F;crates.io&#x2F;crates&#x2F;oxirs-wasm" rel="nofollow">https:&#x2F;&#x2F;crates.io&#x2F;crates&#x2F;oxirs-wasm</a><p>tantivy-wasm: <a href="https:&#x2F;&#x2F;github.com&#x2F;phiresky&#x2F;tantivy-wasm" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;phiresky&#x2F;tantivy-wasm</a><p>Is there an advantage to adding an MCP local memory interface over agent instructions on how to use a rust CLI?<p>And then write Markdown documents with Google OKF-like frontmatter YAML metadata for agents that work with tokens not linked data graphs; <a href="https:&#x2F;&#x2F;github.com&#x2F;GoogleCloudPlatform&#x2F;knowledge-catalog&#x2F;blob&#x2F;main&#x2F;okf&#x2F;SPEC.md" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;GoogleCloudPlatform&#x2F;knowledge-catalog&#x2F;blo...</a>
    • coredog646 hours ago
      Can WASM use AVX512-VNNI?
      • LtdJorge5 hours ago
        No, WASM only has 128b SIMD instructions, for now.
    • cpursley8 hours ago
      Also interested.
  • beernet6 hours ago
    Why not just use Qdrant? They&#x27;ve been integrating TurboQuant for months, works well.
    • kanungle1 hour ago
      Integrated in 5 weeks and just expanded data types for turbo4 in last release. No longer need to store fp32 vectors if you don&#x27;t need them
  • cat-whisperer3 hours ago
    What&#x27;s a good embedding model and search to run locally? something fast and lightweight.
  • sp19828 hours ago
    If anyone is looking to retrofit to an existing pipeline, I use similar ideas to compress vectors for job search, getting roughly 8x compression with about a 3.5% drop in quality. My experiment: <a href="https:&#x2F;&#x2F;corvi.careers&#x2F;blog&#x2F;vector-search-embedding-compression&#x2F;" rel="nofollow">https:&#x2F;&#x2F;corvi.careers&#x2F;blog&#x2F;vector-search-embedding-compressi...</a>
  • OutOfHere3 hours ago
    I am not convinced that Turbovec yields better retrieval than the same amount of bits of a Matryoshka embedding.
  • spoaceman77777 hours ago
    Well. That is insane. O_O Fantastic job!
  • burgerboii8 hours ago
    Who is this co-author called t &lt;t@t&gt;?
    • cute_boi6 hours ago
      As it is heavily vibe coded, I think member of technical staff at antropic has no clue....<p>Next Prompt: remove t@t and force commit.
  • esafak9 hours ago
    lancedb and duckdb integrations would be great...
  • refulgentis7 hours ago
    Bloviating nonsense, 3rd time I’ve seen something like this in HN since TurboQuant came out. You don’t need float32, never did. Source: I’ve been writing on device embedding code for 4 years.
  • zuzululu8 hours ago
    what could i use this for as part of my agentic workflow? codebase indexing? docs ?
    • kyxsc8 hours ago
      notes&#x2F;docs&#x2F;wiki is a great use case
  • cute_boi6 hours ago
    Another vibe coded slop where they can&#x27;t even spend time on Readme or documentation around code...
  • myshapeprotocol3 hours ago
    [dead]
  • tracespect5 hours ago
    [flagged]