Frank Liu
@frankzliu
Followers
577
Following
607
Media
26
Statuses
591
Professional presser of buttons on computer keyboards @VoyageAI @MongoDB
High-dimensional vector space
Joined December 2021
Why pick and choose between MRL and quantization when you can have both? 🤓.
📢 Announcing the new SOTA voyage-3-large embedding model!. • 9.74% over OpenAI and +20.71% over Cohere.• flexible dim. (256-2048) and quantizations (float, int8, binary).• 8.56% over OpenAI with 1/24x storage cost.• 1.16% over OpenAI with 1/192x storage cost ($10K → $52)
3
0
7
RT @MongoDB: Our Multimodal Search Python Library is now in public preview. Giving developers a single interface to build applications tha….
0
5
0
I feel like I've seen this kind of retrieval plot somewhere before 😂.
Introducing Codestral Embed, the new state-of-the-art embedding model for code.
2
1
2
RT @VoyageAI: 📢 Meet voyage-3.5 and voyage-3.5-lite!.• flexible dim. and quantizations.• voyage-3.5 & 3.5-lite (int8, 2048 dim.) are 8% & 6….
0
12
0
RT @continuedev: @metcalfc wrote a deep dive on why your custom AI code assistant should include embeddings and a reranker from @VoyageAI🥇….
0
2
0
RT @Coffee_and_NLP: Among other things that make my day, one of them is a great podcast conversation with my guests. Thanks @frankzliu for….
0
1
0
RT @HanchungLee: @jobergum @JinaAI_ 's definition of deep research is shallow. openai's deep research is a trained system. stanfords storm….
0
1
0
Congrats @nomic_ai! Great to see MoE coming to embedding models.
Nomic Embed Text V2 is now available. - First general purpose Mixture-of-Experts (MoE) embedding model.- SOTA performance on the multilingual MIRACL benchmark for its size.- Support for 100+ languages.- Truly open source - open training data, weights, & code.- Apache 2.0 License
0
1
2
Cosine similarities greater than 1 == "beautiful results" 🤣.
Got some beautiful results with our trial jina-embeddings-v4 checkpoint on modality gap:
1
0
3
RT @flo_re2003: Had a really interesting discussion about agentic retrieval last night at a RAG event at @ExaAILabs with @frankzliu from @V….
0
1
0
RT @michael_chomsky: hyped for the reranking event I'm throwing in sf next thursday:. speakers from:.@ExaAILabs, which just purchased a sup….
0
3
0
RT @TimescaleDB: 🚀 General vs. Domain-Specific: Which Embedding Model Should You Choose for Your RAG App?. We tested OpenAI’s text-embeddin….
0
7
0
voyage-code-3 is one of the first embedding models trained with both Matryoshka learning as well as quantization awareness. More in our blog post:
blog.voyageai.com
TL;DR – Introducing voyage-code-3, our next-generation embedding model optimized for code retrieval. It outperforms OpenAI-v3-large and CodeSage-large by an average of 13.80% and 16.81% on a suite …
📢 Announcing voyage-code-3 embedding model!. 1. more accurate: + 14% gain over OpenAI-v3-large.2. flexible dimension (Matryoshka): 256-2048.3. quantized embeddings: float, int8, binary.4. new Pareto frontier: (binary,256 dim.) is 6% better than OpenAI (float,3072 dim.) 🧵🧵
1
0
6
RT @VoyageAI: Vector-based code retrieval is a critical building block in code assistants and agents. However, many people complained about….
0
14
0
RT @jobergum: I’m glad that there is more interest from embedding providers. Voyager was the first, next up Jina and Nomic? Cohere also.
0
2
0
Recently, I've shared how I believe that "native multimodality" is the future. voyage-multimodal-3, trained end-to-end on text, photos, figures, PDFs, PPTs, and more, is the first embedding model that fits this concept. No more unstructured data ETL. Screenshot is all you need.
📢 Announcing voyage-multimodal-3, our first multimodal embedding model!. It vectorizes interleaved text & images, capturing key visual features from screenshots of PDFs, slides, tables, figures, etc. 19.63% accuracy gain on 3 multimodal retrieval tasks (20 datasets)! 🧵🧵
1
3
35
RT @VoyageAI: Thrilled to share that we've closed $28M in funding, led by @CRV, with continued support from @wing_vc and @saranormous. Also….
voyageai.com
Voyage AI provides cutting-edge embedding models and rerankers for search and retrieval
0
35
0