fzliu

41 ms·

721.
by fzliu 3y ago
Hola Jeff. Let's host a meetup together.
722.
Vector Database in a Jupyter Notebook(zilliz.com)
72 points by fzliu 3y ago|6 comments
723.
Chat with Open Source Software(osschat.io)
3 points by fzliu 3y ago|0 comments
724.
HNSW for Vector Search(zilliz.com)
1 points by fzliu 3y ago|0 comments
725.
Render a neural network into CUDA/HIP code(github.com)
177 points by fzliu 3y ago|65 comments
726.
Evaluate Multiple LLMs Easily(github.com)
1 points by fzliu 3y ago|0 comments
727.
Hierarchical Navigable Small Worlds(zilliz.com)
1 points by fzliu 3y ago|0 comments
728.
Hierarchical Navigable Small Worlds(zilliz.com)
2 points by fzliu 3y ago|0 comments
729.
by fzliu 3y ago
Transformers don't seem to optimal as an architecture for vision tasks, but I don't think pure convnets are either. Either way, having the right data is 10x as important.
730.
Future of NLG evaluation: LLMs and high quality human eval?(ehudreiter.com)
1 points by fzliu 3y ago|0 comments
731.
Introduction to Vector Similarity Search(zilliz.com)
1 points by fzliu 3y ago|0 comments
732.
ML Observability in a Notebook(github.com)
2 points by fzliu 3y ago|0 comments
733.
by fzliu 3y ago
Here's one that may be helpful for you: https://zilliz.com/blog/using-vector-database-search-white-h...
734.
Chat with Open Source Software(osschat.io)
2 points by fzliu 3y ago|0 comments
735.
A Teenager's Guide to Avoiding Actual Work (2021)(madned.substack.com)
1 points by fzliu 3y ago|0 comments
736.
Observability for Generative AI(helicone.ai)
2 points by fzliu 3y ago|0 comments
737.
by fzliu 3y ago
AFAIK pgvector only supports IVF, no PQ (nor HNSW).
738.
by fzliu 3y ago
A naive PQ index that trains the quantizer once and reuses the same codebook for recent data definitely has this problem. We (Milvus/Zilliz) break up all the vectors into multiple indexes at certain checkpoints which we call segments.
739.
by fzliu 3y ago
Just read the abstract and intro for the link you posted - thanks for sharing. As great as HNSW and other graph-based indexes are, I think PQ and other encoder/decoder-based methods are still incredibly important for ANN search in gene
740.
by fzliu 3y ago
For folks interested in the rest of our vector search 101 series, here's a link: https://zilliz.com/blog?tag=39&page=1
741.
Product quantization for vector search(zilliz.com)
51 points by fzliu 3y ago|8 comments
742.
Product Quantization(zilliz.com)
1 points by fzliu 3y ago|0 comments
743.
by fzliu 3y ago
AFAIK for over-parameterized models, performing quantization or any other form of compression won't reduce accuracy by much (don't quote me on this though).
744.
Prompt Injection Detector(github.com)
1 points by fzliu 3y ago|0 comments
745.
by fzliu 3y ago
I'm also not entirely convinced by "huge" context models just yet, especially as it relates to fuzzy knowledge such as overarching themes or writing style. In particular, there are 0 mentions of the phrase "machine learn
746.
by fzliu 3y ago
I don't understand how this can be considered a technical report. No information on model architecture, distributed training methodology, or optimizations. The "Training dataset" section is a pathetic 0.5 pages long. Come on,
747.
by fzliu 3y ago
Not sure how you arrived at this conclusion. Mouser and Digi-Key are nearly identical in terms of UI. Search functionality on top followed by products & manufacturers and a list of products on the left. I'm with you on that second
748.
by fzliu 3y ago
Most commonly used ANN algorithms today are clever optimizations atop "old" data structures. DiskANN is a recent favorite: https://milvus.io/blog/2021-09-24-diskann.md
749.
by fzliu 3y ago
Totally agree. It takes _a lot_ to go from Hnswlib to a full-fledged vector database. Here's an architecture diagram for a production-ready vector database https://milvus.io/docs/architecture_overview.md . Not exac
750.
Rufus: The Reliable USB Formatting Utility(github.com)
2 points by fzliu 3y ago|1 comments

More ›