StringZilla

ined on balanced multilingual datasets, with equal exposure to languages from English, Chinese, and Hindi to Arabic, Hebrew, and Armenian. UForm now supports 21 languages, is so tiny that you can run it in the browser, and outputs small 256-dimensional embeddings. Perfect for rapid image and video searches. It's already available on Hugging-Face as "unum-cloud/uform-vl-multilingual-v2". UCall [3]: It started as a FastAPI alternative focusing on JSON-RPC (instead of REST protocols), offering 70x the bandwidth and 1/50th the latency. It was good but not enough, so we've added REST and TLS support, broadening its appeal. I've merged that code, and it is yet to be tested. Early benchmarks suggest that we still hit the same 150'000-250'000 requests/s on a single CPU core in Python by reusing HTTPS connections. StringZilla [4]: This project lets you sift through multi-gigabyte or terabyte strings with minimal use of RAM and maximal use of SIMD and SWAR techniques. All these projects are engineered for scalability and efficiency, even on tight budgets. Our demo, for instance, works on hundreds of gigabytes of images using just a few gigabytes of RAM and no GPUs for AI inference. That is a toy example with a small, noisy dataset, and I look forward to showing a much larger setup. Interestingly, even this tiny setup illustrates issues common to UForm and much larger OpenAI CLIP models - the quality of Multi-Modal alignment [5]. It also shows how different/accurate the search results are across different languages. Synthetic benchmarks suggest massive improvements for some low-resource languages (like Armenian and Hebrew) and more popular ones (like Hindi and Arabic) [6]. Still, when we look at visual demos like this, I can see a long road ahead for us and the broader industry, making LLMs Multi-Modal in 2024 :) All of the projects and the demo code are available under an Apache license, so feel free to use them in your commercial projects :) PS: The demo looks much nicer with just Unsplash dataset of 25'000 images, but it's less representative of modern AI datasets, too small, and may not be the best way to honestly show our current weaknesses. The second dataset - Conceptual Captions - is much noisier, and quite ugly. [1]: https://github.com/unum-cloud/usearch [2]: https://github.com/unum-cloud/uform [3]: https://github.com/unum-cloud/ucall [4]: https://github.com/ashvardanian/stringzilla [5]: https://ashvardanian.com/posts/searching-stable-marriages/#e... [6]: https://www.unum.cloud/blog/2023-08-17-uform-graphcore

o very very badly for longer needles. * It seems quite likely that your [confirmation step](https://github.com/ashvardanian/Stringzilla/blob/fab854dc4fd...) is going to absolutely kill performance for even semi-frequently occurring candidates. The [`memchr` crate utilizes information from the vector step to limit where and when it calls `memcmp`](https://github.com/BurntSushi/memchr/blob/46620054ff25b16d22...). Your code might do well in cases where matches are very rare. I took a quick peek at your benchmarks and don't see anything that obviously stresses this particular case. For substring search, the `memchr` crate uses a variant of the "[generic SIMD](http://0x80.pl/articles/simd-strfind.html#first-and-last)" algorithm. Basically, it takes two bytes from the needle, looks for positions where those occur and then attempts to check whether that position corresponds to a match. It looks like your technique uses the first 4 bytes. I suspect that might be overkill. (I did try using 3 bytes from the needle and found that it was a bit slower in some cases.) That is, two bytes is usually enough predictive power to lower the false positive rate enough. Of course, one can write pathological inputs that cause either one to do better than the other. (The `memchr` crat benchmark suite has a [collection of pathological inputs](https://github.com/BurntSushi/memchr/blob/46620054ff25b16d22...).) It would actually be possible to hook Stringzilla up to `memchr`'s benchmark suite if you were interested. :-) [1]: https://old.reddit.com/r/rust/comments/163ph8r/memchr_26_now... [2]: https://github.com/BurntSushi/memchr