we could easily add that if there's interest, e.g. using Pub/Sub and Cloud Storage. If there are .NET libraries, should be straightforward implementing some interfaces.
Similar considerations for the inference part, embedding and text generation.
For instance, given the user "ask" (which could be any generic message in a copilot), decide how to query one or multiple storages. Ultimately, companies and users have different storages, and a few can be indexed with vectors (and additional fine tuned models). But there's a lot of "legacy" structured data accessible only with SQL and similar languages, so a "planner" (in the SK sense of planners) could be useful to query vector indexes, text indexes and knowledge graphs, combining the result.
Currently we use LLMs to generate a summary, used as an additional chunk. As you might guess, this can take time, so we postpone the summarization at the end (the current default pipeline is: extract, partition, gen embedding, save embeddings, summarize, gen embeddings (of the summary), save emb)
Initial tests though are showing that summaries are affecting the quality of answers, so we'll probably remove it from the default flow and use it only for specific data types (e.g. chat logs).
There's a bunch of synthetic data scenarios we want to leverage LLMs for. Without going too much into details, sometimes "reading between the lines", and for some memory consolidation patterns (e.g. a "dream phase"), etc.
We are also developing an open-source solution for those who would like to test it out and/or contribute, it can be consumed as a web service, or embedded into .NET apps. The project is codenamed "Semantic Memory" (available in GitHub) and offers customizable external dependencies, such as using Azure Queues, RabbitMQ, or other alternatives, and options for Azure Cognitive Search, Qdrant (with plans to include Weaviate and more). The architecture is similar, with queues and pipelines.
We believe that enabling custom dependencies and logic, as well as the ability to add/remove pipeline steps, is crucial. As of now, there is no definitive answer to the best chunk size or embedding model, so our project aims to provide the flexibility to inject and replace components and pipeline behavior.
Regarding Scalability, LLM text generators and GPUs remain a limiting factor also in this area, LLMs hold great potential for analyzing input data, and I believe the focus should be less on the speed of queues and storage and more on finding the optimal way to integrate LLMs into these pipelines.
The model doesn't run code, it generates text that happens to be code. It's up to the client calling the model API to use this text, e.g. compiling and executing (if that's your scenario) and calling the model again to fix the original code if needed.
the search filter seems to have some hard coded logic, preferring languages like R (4 projects [1]) over Scala (5 projects [2]). I wonder if that's based on trends or purely taste (or just cache :-)).
Having worked mostly with mesos and k8s, I found k8s configuration superior, e.g. how one can import secrets, config files, and more importantly set up the network without address translation or port forwarding.
Tooling seems OK with both, and I agree one needs to spend some time to get familiar with the CLI and nomenclature, IMHO because both are quite flexible and powerful.
Trying the site with chrome on android, some frustrating issues:
* as soon as logos appear the signup form covers all of them, so I can't see the logos. Is it by design? I tried signing up and the wheel keeps spinning, I'll retry later.
* Since I tried multiple times, the browser allows to autofill forms. When I do that the js validation fails thinking I haven't entered any char yet.
* not sure what to do on the Selectr-icons-screen, I've tried clicking here and there with no results. Luckily I can skip the step :-)
* Reloading the page (eg I'm wondering if the server is overloaded and I want to give it a chance) the logo generation restart from scratch. Pls consider keeping state in the URL.
Looks like the announcement is still under work, e.g. you can visit 2.12.0-RC1 page [1] and 2.12.0-RC2 [2] but not 2.12.0 [3] yet, and both http://get-scala.org and http://scala-lang.org homepages point to 2.11.8 as the official version.
BTW, here's a list of packages available for 2.12.x [4], if you are planning to migrate (please do!). The package is on Maven [5] :-)
From the release notes [6] "This release is identical to 2.12.0-RC2" so anything on RC2 should just work.
BTW "2.12.1 will be out shortly (by the end of November) to address some known (but rare) issues in 2.12.0, and 2.11.9 is the last planned 2.11.x release (due by mid December). In the next few weeks, we at Lightbend will share our plans for Scala 2.13." [4]
@soared, sorry if I gave that impression. I'm actually a software engineer, enthusiast HN reader. I thought this was a good opportunity to join the community and share some work :-)