Assuming that's what he meant, why would it be considered baseline versus anything else? I am genuinely curious because I'd like to know more about issues people face with this or that vector store in general.
I am working on search but rather for text-to-image retrieval, nevertheless, I am curious if by that's all baked into Azure AI search you also meant synthetic query generation from the grandparent comment. If so, what's your latency for this? And do you extract structured data from the query? If so, do you use LLMs for that?
Moreover I am curious why you guys use bm25 over SPLADE?
Yes, AI Search has a new agentic retrieval feature that includes synthetic query generation: https://techcommunity.microsoft.com/blog/azure-ai-foundry-bl...
You can customize the model used and the max # of queries to generate, so latency depends on those factors, plus the length of the conversation history passed in. The model is usually gpt-4o or gpt-4.1 or the -mini of those, so it's the standard latency for those.
A more recent version of that feature also uses the LLM to dynamically decide which of several indices to query, and executes the searches in parallel.
That query generation approach does not extract structured data. I do maintain another RAG template for PostgreSQL that uses function calling to turn the query into a structured query, such that I can construct SQL filters dynamically.
Docs here:
https://github.com/Azure-Samples/rag-postgres-openai-python/...
Got it, I think this might make sense for a "conversation" type of search not for an instant search feature because lowest latency is gonna be too high IMO.
Fair point on latency, we (Azure AI Search) target both scenarios with different features. For instant search you can just do the usual hybrid + rerank combo, or if you want query rewriting to improve user queries, you can enable QR at a moderate latency hit. We evaluated this approach at length here: https://techcommunity.microsoft.com/blog/azure-ai-foundry-bl...
Of course, agentic retrieval is just better quality-wise for a broader set of scenarios, usual quality-latency trade-off.
We don't do SPLADE today. We've explored it and may get back to it at some point, but we ended up investing more on reranking to boost precision, we've found we have fewer challenges on the recall side.
I’ll admit I haven’t checked before posting, perhaps an admin can merge both submissions and change the URL on the one you linked to the one in this submission.
In the case where you still have an image locally, trying to build will fail with an error complaining about not being able to load metadata for the image because a HEAD request failed. So, the real question is, why isn't there a way to disable the HEAD request for loading metadata for images? Perhaps there's a way and I don't know it.
Unfortunately, it seems like it's not working properly anymore. I just messaged Bret Victor and maybe he can get it back in working order or can reach out to someone.
> vision-based search for comprehensive document understanding
but it's not clear to me what this means, is it just vector embeddings for each image in every document via a CLIP-like model?
In addition, I'd be curious what's the rationale behind using the plethora of databases, given the docs on running it in production spins them all up, I assume they're all required, for instance I'd be curious on the trade-offs between using postgres with something like pg_search (for bm25 support, which vanilla postgres FTS doesn't have) vs using both postgres and ElasticSearch.
The docs are also very minimal, I'd have loved to see at least 1 example of usage.
This is indeed one of the major differences. Many of the problems that are plaguing Vispy are related to OpenGL. The use of wgpu solves many of them.
Also, wgpu forces you to prepare visualizations in pipeline objects, which at drawtime require just a few calls. In OpenGL there is way more work for each object being visualized at drawtime. This overhead is particularly bad on Python. So this particular advantage of wgpu is extra advantageous for Python.
A while back I wrote this https://mohamed.computer/posts/python-internals-cpython-byte..., perhaps it's interesting for people who use `dir` and wonder what some of the weird things that show up are.