vespa.ai @vespaengine
https://t.co/abkb8IjPSH - the open source platform for combining data and AI, online. Vectors/tensors, full-text, structured data; ML model inference at scale. vespa.ai Joined September 2017-
Tweets416
-
Followers3K
-
Following5
-
Likes547
Having trouble keeping up? Guidebook to the State-of-the-Art Embeddings and Information Retrieval by @aapo_tanskanen at @thoughworks is out today - a great resource to get up to date. linkedin.com/pulse/guideboo…
This is what customer obsession looks like. Props to @vespaengine team for promoting what is effectively 40x cheaper for the user.
This is what customer obsession looks like. Props to @vespaengine team for promoting what is effectively 40x cheaper for the user.
If you are using vector embeddings, reading this post might be the most profitable ten minutes you'll ever spend.
If you are using vector embeddings, reading this post might be the most profitable ten minutes you'll ever spend.
Matryoshka 🤝 Binary vectors: Slash vector search costs with Vespa We announce support for combining matryoshka and binary quantization in Vespa’s native hugging-face embedder and discuss how this slashes vector search costs. blog.vespa.ai/combining-matr…
The latest Vespa newsletter is here to help you stay up to date on what's happening on the leading edge in RAG, IR and vector search: - A new SPLADE embedder - ONNX models with float16 - @cohere embedding model guides - Support for an array of chunks with ColBERT - And list of…
If you are in Paris tonight, you should check out this meetup with @kraune from @vespaengine aicamp.ai/event/eventdet…
Seems everybody is migrating their search and recommendation systems from Elastic to Vespa now. Here's the experience of Stanby, Japan's leading job search site: blog.vespa.ai/migrating-to-t…
Happy binary text embedding week! Created a quick notebook demonstrating: - Using @mixedbreadai embed-large-v1 model with the new sentence-transformer API for - How to index binary embeddings with HNSW indexing in @vespaengine - float-binary re-ranking! pyvespa.readthedocs.io/en/latest/exam…
Happy binary text embedding week! Created a quick notebook demonstrating: - Using @mixedbreadai embed-large-v1 model with the new sentence-transformer API for - How to index binary embeddings with HNSW indexing in @vespaengine - float-binary re-ranking! pyvespa.readthedocs.io/en/latest/exam… https://t.co/91yB2H3BtM
A new Vespa sample app is out, featuring the brand new native Vespa splade embedder. Thank you for open-sourcing the sparse encoder model @prithivida and to @NirantK for uploading to HF! github.com/vespa-engine/s… search.vespa.ai/search?query=W…
This is an example of a phased coarse-to-fine ranking pipeline. This you can only do with @vespaengine. Nils also touches on the limitations of default provisioning settings of EBS volumes. An insightful thread on both performance and storage-tier economics for vector search.
This is an example of a phased coarse-to-fine ranking pipeline. This you can only do with @vespaengine. Nils also touches on the limitations of default provisioning settings of EBS volumes. An insightful thread on both performance and storage-tier economics for vector search. https://t.co/2pm3OHWJsI
Binary embeddings from @cohere with @vespaengine! - HNSW index with hamming distance over 1024 bits! - Re-ranking with the dot product between full query vector (1024 floats) against an unpacked float version of the binary embedding. Notebook: pyvespa.readthedocs.io/en/latest/exam…
When GigaOm named Vespa Leader in their Sonar for Vector Databases, one of the categories where we scored Excellent were Embedding Flexibility - why? Vespa lets you create embeddings in four ways: - On your own, outside Vespa: Just pass tensors directly in documents and…
Talking about single-vector databases, for the 200K long-document MLDR dataset we store 614M vectors on a single node for late context level interaction or late cross-context interaction. Tensors is the way blog.vespa.ai/announcing-lon…
Jo Kristian Bergum @jobergum
9K Followers 814 Following Distinguished Engineer @vespaengine. Tweets about Vespa, search, recommendation, ranking, and IR. CET. #StandWithUkraine 💙💛
Nils Reimers @Nils_Reimers
10K Followers 434 Following Director of Machine Learning @Cohere | ex-huggingface | Creator of SBERT (https://t.co/MKKOMfuQ4C)
Jeremy Howard @jeremyphoward
222K Followers 5K Following 🇦🇺 Co-founder: @AnswerDotAI & @FastDotAI ; Hon Professor: @UQSchoolITEE ; Digital Fellow: @Stanford
Jay Alammar @JayAlammar
35K Followers 1K Following Machine learning and language models R&D. Builder. Writer. Visualizing AI, ML, and LLMs one concept at a time. @Cohere. https://t.co/TquuQXlLOJ
Jimmy Lin @lintool
13K Followers 842 Following I profess CS-ly at the @UWaterloo and gaze into the technological crystal ball at @Primal. I used to write code for @Twitter and slides for @Cloudera.
Charlie Hull @FlaxSearch
2K Followers 860 Following Managing Consultant at OpenSource Connections, helping you build amazing AI & search applications. Also hachyderm dot io slash @flaxsearch
Doug Turnbull @softwaredoug
3K Followers 754 Following Search @Reddit; ex @Shopify & @o19s; Books: Relevant Search & AI Powered Search
Charly Wargnier @DataChaz
112K Followers 31K Following 🥑 DevRel @Streamlit @SnowflakeDB 🪶 𝕏 about #AI, #LLMs, #DataScience, #WebApps, #SEO 💕 My heart is open source 🌍 Nature Lover 👀 My views!
Alexander Reelsen @spinscale
3K Followers 1K Following Husband, dad, enjoys working distributed, likes distributed systems, data stores, JVM/Java & Basketball/Streetball, now at https://t.co/YPftyN9nyB
Simon Willison @simonw
71K Followers 5K Following Creator @datasetteproj, co-creator Django. PSF board. @nichemuseums. Hangs out with @natbat + @cleopaws. He/Him. Mastodon: https://t.co/t0MrmnJW0K
Sean MacAvaney @macavaney
1K Followers 479 Following he/him · Lecturer (Assistant Professor) at @GlasgowCS @TerrierTeam · working at the intersection of IR&NLP · PhD from @Georgetown IRLab Website: https://t.co/TvZBNq61Ey
Dan(i(el(e))) −·�.. @LelViLamp
8 Followers 86 Following One day I'd like to open a shop in London specialising in pirate memory games 🧩 I speak 🥨💂♂️🤌🥖🍣🐻 .NET dev 💻 PG Data Science 🧮 linguist by heart 🌈🇪🇺
pierrix00 @pierrix00
67 Followers 191 Following
Vox - e/acc @TheVoxxx
586 Followers 175 Following
Wilfred (Willem) Mijn.. @wmijnhardt
2K Followers 3K Following Policy Director General @RSMErasmus & Honorary Prof. @EBS_Global HWU. Passionate for Universities, Business Schools, #excellence #impact #RRI #RRBM #PRME
Rasmus Toivanen @RasmusToivanen
655 Followers 2K Following More generalist than specialist. Industrial engineer turned into ML/DL. Currently DE @RecordlyData by day, Applied DS research by night with @aapo_tanskanen
Patrick Clear @PatrickClear8
8 Followers 42 Following
Abdulrahman Tabaza @embed_dim
4 Followers 799 Following enjoyer of various vector spaces, encoders and modalities
Aakarsh Ramchandani @2sidesofacoin
99 Followers 529 Following Running strategy @ravenpack. Board member @ Poocho. Previously @thirdpointllc and @factset. Currently obessing over domain-spec LLMs & vectorizing everything
SenaBeren @findingmerit
287 Followers 3K Following
Amir Soleimani @ASoleimaniB
233 Followers 253 Following AI Engineer at Sdu. Generative AI for Legal Research
Billy Vythikowski @vythikowski
29 Followers 317 Following
Pratheek Rebala @pratheekrebala
922 Followers 2K Following News developer @publicintegrity. Shop steward @publici_union.
Anthony Bordonaro @bordo_anthony
454 Followers 1K Following Engineering Manager, @carltonfc supporter, green tea enthusiast
Leonce Nshuti @LeonceNshuti
278 Followers 2K Following Data Engineer @Sony. Ex-UBS, Vanderbilt, Harvard. https://t.co/kOPPM3IA54. Google Scholar: https://t.co/UWXNmktdq0. Opinions my own.
sportscarfan45 @unsafetensors
50 Followers 651 Following
TENTANANO @tentanano
7 Followers 125 Following
Karthic Natarajan @karthicn_
377 Followers 2K Following Cloud & Data Analytics | Product Developer | Tech Enthusiast | Interests - System Design & Architecture | Learning- Minimalism | 5x OCI & 4x Azure Cloud
Mark R. Hinkle @mrhinkle
7K Followers 5K Following I help enterprises understand and use artificial intelligence. Leveraging my 25 years of enterprise software experience in emerging technology to drive results.
YChu.eth @ychudoteth
365 Followers 2K Following Invent the decentralized future with love! | Prev: Engineer at ByteDance
Romain Damery @RomainDamery
2K Followers 2K Following Leading technical SEO and other nerdy stuff @AmsiveAgency / ((bb) || !(bb)) / 🇺🇸 🇫🇷 🇪🇺 🇧🇷 🏳️🌈
Ian Maurer 🧬🤖�.. @imaurer
1K Followers 793 Following CTO @GenomOncology #genomics #precisiononcology #nlp
Sayan Chakraborty @shockrobortyy
150 Followers 905 Following ML @Qualcomm (prev: @BrownUniversity, @paytminsider, @bigbinary, @clarisights)
Elorm Dokosi @ElormDokosi
123 Followers 561 Following Computer science and engineering student. Learning to be an indie hacker on the side.
dzh886 @dengzihao88
22 Followers 586 Following
Gordon Lindsay (Busin.. @GLbusiness_twit
1 Followers 4 Following
Jon Page @jonpage0
71 Followers 403 Following
CryptoLaika @DemonsZzh
126 Followers 1K Following
Consiglieri @ConsiglieriVita
84 Followers 2K Following
Vlado Handziski @vlahan
980 Followers 5K Following Unwiring the future as CTO of @R3Coms | #Wireless #IIoT | Ex academic @TUBerlin | Husband and father of two | Opinions my own | https://t.co/32UvARYu5g
fraserxu @fraserxu
624 Followers 398 Following Lead engineer @envato, past organiser of @jsconfchina and Shanghai JavaScript meetup.
Tomasz Kobylinski @TmoaszKo
10 Followers 42 Following
Jack&Penny @Jack870202
34 Followers 246 Following Computer Engineer, Texas Hold'em Lover and Crazy Sports fans
Jimmy Sticks @loss_gobbler
128 Followers 423 Following Chief Symbiosis Sorcerer at @stickshiftAI ////////////////// lair dweller // FAANG quitter // cyborgism enjoyer
tm @tm57312196
14 Followers 187 Following Interest: physical understanding of consciousness, wisdom etc.
FluffyVectors ☁️�.. @FluffyVectors
7 Followers 88 Following The AI context platform for everyone. Memories 🧠, Preferences 👍, Semantic Search 🔍 (and more) for AI Models, Agents, and Multi-Agent systems #fluffyvectors
WΞNDΞL @0xwendel
2K Followers 482 Following Attention Token Engineer, Medical LLM Inference OP and sometimes Solidity Dev, https://t.co/yRPtosJ9Hi
うりんつ@QA @yurinzflet
12 Followers 83 Following Software QA. Former Medical device QA. IT未経験QA立ち上げ → そのままマネージャー3年目。Playwrightが好き。もっとコーディングがしたいお年頃 All tweets are on my own
Robert @clarity99
500 Followers 970 Following gestalt psychotherapist, mindfulness teacher and an all around geek. ;) Moved to Mastodon: https://t.co/Qzlp34vmWV
takono @takono0807
112 Followers 325 Following Web系ソフトウェアエンジニアです。 楽しいチームでいい感じのプロダクトをつくりたい。 技術以上に人間力を身につけたい30代。
Georgi D. Gospodinov @ggospodi
258 Followers 872 Following Technologist and entrepreneur PhD in math https://t.co/Wb6iH1UClJ
Black Birkin @iidecat
0 Followers 427 Following
Jo Kristian Bergum @jobergum
9K Followers 814 Following Distinguished Engineer @vespaengine. Tweets about Vespa, search, recommendation, ranking, and IR. CET. #StandWithUkraine 💙💛
Sarah Catanzaro @sarahcat21
12K Followers 1K Following “All methods are sacred if they are internally necessary” (GP @amplifypartners, prev @canvasvc; Head of Data @Mattermark; @palantirtech; @c4ads)
Thiago Guerrera @Thiagogm
5K Followers 139 Following Working on https://t.co/kTksMcNTfG. Statistics is my craft.
Jon Bratseth @jonbratseth
358 Followers 47 Following CEO https://t.co/5qXgcEp1MU Build things and help people.Three things you should know about scaling embedding retrieval systems: - Embedding dimensionality cost: linear with dims - Embedding inference cost: quadratic with tokens - ANN (HNSW) cost: sub-linear with documents
A 20-page book (+ charts and tables) guidebook to state-of-the-art embeddings and information retrieval. linkedin.com/pulse/guideboo…
Looking at the MTEB leaderboard this AM. Amazingly, mxbai-embed-large-v1 ranks at 12 despite its small size relative to the other B-parameters models. In addition to strong performance for a relatively small size, it comes with MRL and BQL flexibility. blog.vespa.ai/combining-matr…
Hamming distance got a nice speedup in Vespa this week, 38% faster. It is approaching 1 billion 64-dimensional int8 hamming distances per second on a single CPU socket, 20x faster than the normalized dot product (1024-dim). Exact nearest neighbor search.
I'll be discussing MRL, including recent developments from @OpenAI embedding models and excellent work built on MRL from folks at @vespaengine , @nomic_ai, Sentence Transformers (@tomaarsen) @supabase , to name a few. 2/n
Random observation, but we at @vespaengine have been used to running ML workloads on Vespa with deep-learned embeddings in production at scale for more than a decade.
This is what customer obsession looks like. Props to @vespaengine team for promoting what is effectively 40x cheaper for the user.
The emphasis of both MRL and BQL is on sacrificing accuracy by a few % in exchange for a much lower cost. By using a compact representation of the text embeddings, the systems can run on less expensive hardware or require less memory, resulting in cost savings.
I keep having to re-check my calculations. Can I really fit decent quality embeddings for my 40 million docs into a few GB of ram?! Vespa integrating MRL + binarization + bit packing for a huge win 🔥🔥 Looking forward to working through this post and trying it out on my data.…
Matryoshka 🤝 Binary vectors: Slash vector search costs with Vespa We announce support for combining matryoshka and binary quantization in Vespa’s native hugging-face embedder and discuss how this slashes vector search costs. blog.vespa.ai/combining-matr…
After 3 weeks at @vespaengine, I still get more and more impressed every day by both the engine and the team 🤩 Check out this blog post to explore some features that sets it apart!
Matryoshka 🤝 Binary vectors: Slash vector search costs with Vespa We announce support for combining matryoshka and binary quantization in Vespa’s native hugging-face embedder and discuss how this slashes vector search costs. blog.vespa.ai/combining-matr…
We are very excited about this direction, as it unlocks many new use cases that are no longer prohibitively expensive to serve in production—making more unstructured data useful. The new Vespa embedder functionality for MRL and BQ is available in Vespa 8.332.5 and above. Enjoy
The unique aspect of MRL and BQL is that they introduce minimal computational overhead during embedding model training. Both techniques are post-processing steps performed after model inference.
Approaching 10,000 queries per second with 100,000 vectors equals close to 1B hamming distance computations per second! This is on a single CPU.
Calculating the hamming distance is approximately 20 times faster (2ms), enabling users to experience faster search and higher query throughput with the same resources. In practical terms, organizations can reduce CPU-related costs by 20x
With BQL (binary quantization) to binary vectors, we both get 32x less memory resource footprint but also a significant speedup over the float representations.
On serving performance, MRL gives us a linear reduction in cost. The graph above demonstrates that reducing the float dimensions via MRL from 1024 to 512 results in a two-fold speedup.
The representations can be used in a phased retrieval and ranking pipeline in Vespa
With the announced Vespa hugging-face-embedder support, developers can easily obtain multiple vector representations with a single inference call
.@mixedbreadai's combination of MRL and BQL retains 90% accuracy (on MTEB Retrieval task) using 64-dimensional int8 (512 bits) binarized from 512 float dimensions (first 512 out of 1024). This binary representation reduces storage-related costs by 64 compared to the baseline
Since both techniques are simple post-processing steps over the embedding vector representation, we can produce multiple representations with a single model inference call. One inference pass is vital because model inference is a significant cost driver for embedding retrieval
The emphasis of both MRL and BQL is on sacrificing accuracy by a few % in exchange for a much lower cost. By using a compact representation of the text embeddings, the systems can run on less expensive hardware or require less memory, resulting in cost savings.
Trends for United States
You might like
