ashvardanian

10 ms·

1. 151.▲
   by ashvardanian 1y ago

   Qwen models have historically been pretty good, but there seems to be no architectural novelty here, if I’m not missing it. Seems like another vision encoder, with a projection, and a large autoregressive model. Have there been any better i
2. 152.▲
   Archestra – open-source MCP orchestrator for everyone(github.com)

   25 points by ashvardanian 1y ago|4 comments
3. 153.▲
   by ashvardanian 1y ago

   Thanks for the kind words! In this case, it isn’t tied to any programming language or locale-specific formatting. I just find commas less readable in long numbers, especially in running text across Western languages. Apostrophes feel cleare
4. 154.▲
   by ashvardanian 1y ago

   Thanks! You are likely right! It took a lot of time to make sure that all 6 of ISA-specific versions of StringZilla ( https://github.com/ashvardanian/StringZilla/blob/main/includ... ) return the same outpu
5. 155.▲
   by ashvardanian 1y ago

   Yes, at the scale of 128-bit registers NEON is mostly enough, except for a few categories of instructions missing in that ISA subset, like scatter/gather ops, that can yield 30% boost over serial memory accesses: https://git
6. 156.▲
   Lessons from leaders who turned AI challenges into wins(fastcompany.com)

   1 points by ashvardanian 1y ago|0 comments
7. 157.▲
   by ashvardanian 1y ago

   The consumer situation is changing. A few years ago, when I was working with a team on some closed source HPC stuff, we’ve got everyone Tiger Lake-based laptops to simplify AVX-512 R&D. Now, Zen4-based desktop CPUs also support it. But
8. 158.▲
   by ashvardanian 1y ago

   Not that I’m aware of. Some commercial DBMS vendors are experimenting with integrations, but I haven’t really seen much in the Postgres ecosystem. What excites me in this release is the quality of the new hash functions. I’ve built many ove
9. 159.▲
   by ashvardanian 1y ago

   This last wave of work was actually triggered by the industry over the last 2 years, as the volume of biological sequence data is growing rapidly and more BioTech and Pharma companies are rushing to scale computational pipelines. Namely, if
10. 160.▲
    by ashvardanian 1y ago

    Makes sense! I mostly focus on newer AVX-512 variants as opposed to older AVX2-only CPUs. As for aarch64, it is supported with both NEON, SVE, and SVE2 kernels for some tasks. The last two are rarely useful, unless you run on AWS Graviton 3
11. 161.▲
    by ashvardanian 1y ago

    It’s a very nice and detailed benchmark suite! Great effort! Can you please share the CPU model you are running on? I suspect it’s an x86 CPU without AVX-512 support.
12. 162.▲
    by ashvardanian 1y ago

    I’m not an active RipGrep user, so can’t speak for all usage patterns. My guess: for plain substring searches, you probably won’t see much difference. Where StringZilla may potentially help is in character-set searches.
13. 163.▲
    by ashvardanian 1y ago

    After publishing this a few days ago, 2 things have happened. First, it tuned out that StringZilla scales further to over 900 GigaCUPS around 1000-byte long inputs on Nvidia H100. Moreover, the same performance is obviously accessible on lo
14. 164.▲
    Taking a Look at Compression Algorithms(cefboud.com)

    8 points by ashvardanian 1y ago|0 comments
15. 165.▲
    Processing Strings 109x Faster Than Nvidia on H100(ashvardanian.com)

    216 points by ashvardanian 1y ago|26 comments
16. 166.▲
    Why Do LLMs Design Mediocre Architecture?(recurse.ml)

    2 points by ashvardanian 1y ago|2 comments
17. 167.▲
    Zencoder Lets Developers Bring Their CLI Coding Agent of Choice to Its Platform(thenewstack.io)

    1 points by ashvardanian 1y ago|0 comments
18. 168.▲
    Proprietary sector indexes across tech and beyond(multiples.vc)

    1 points by ashvardanian 1y ago|0 comments
19. 169.▲
    The Constexpr Debugger(blog.jetbrains.com)

    2 points by ashvardanian 1y ago|0 comments
20. 170.▲
    by ashvardanian 1y ago

    Largely true, but there is always a caveat. On Zen4 and Zen4c the register is 512 bits wide. However, internally, many “datapaths” (execution units, floating-point units, vector ALUs, etc.) are 256 bits wide for much of the AVX-512 function
21. 171.▲
    by ashvardanian 1y ago

    My expectation, they will perform great! I’m now mostly benchmarking on 192 core Intel, AMD, and Arm instances on AWS, and in some workloads they come surprisingly close to GPUs even on GPU-friendly workloads, once you get the SIMD and NUMA
22. 172.▲
    by ashvardanian 1y ago

    Sadly, no! On the bright side, they support new AVX2 VNNI extensions, that help with low precision integer dot products for Vector Search! SimSIMD (inside USearch (inside ClickHouse)) already has those SIMD kernels, but I don’t yet have the
23. 173.▲
    Optimizing ClickHouse for Intel's ultra-high core count processors(clickhouse.com)

    225 points by ashvardanian 1y ago|51 comments
24. 174.▲
    {fmt} v12: 60% faster double formatting, full constexpr, and C++ 20 modules(github.com)

    2 points by ashvardanian 1y ago|0 comments
25. 175.▲
    by ashvardanian 1y ago

    For Vector Search the top 2 are: Meta’s FAISS and (my) Unum’s USearch. Lucene powers Elastic, Solr, MongoDB Atlas, AWS OpenSearch, Azure Cognitive Search. USearch powers ClickHouse, DuckDB, YugaByte, TiDB, ScyllaDB, MemGraph, KuzuDB, Lanter
26. 176.▲
    by ashvardanian 1y ago

    Nice observation! Took me a while to realize that Grace Blackwell refers to a person and not an Nvidia chip :) I’ve worked with large genomic datasets on my own dime, and the default formats show their limits quickly. With FASTA, the first
27. 177.▲
    Fun with Google Scholar(diffuse.one)

    1 points by ashvardanian 1y ago|0 comments
28. 178.▲
    by ashvardanian 1y ago

    Lucene is tough to deal with. About 15 hours ago — right when this comment was posted — I was giving a talk at Databricks comparing the world’s most widely used search engines. I’ve never run into as many issues with any other similar tool
29. 179.▲
    by ashvardanian 1y ago

    The question is how many, and what kind of VMs you use? It greatly affects performance :) I run a lot of search-related benchmarks ( https://github.com/ashvardanian ) and curious if you’ve compared to other engines on the sam
30. 180.▲
    by ashvardanian 1y ago

    Very curious about the hardware setup used for this benchmark!