michaelneale

@null not as yet, but did a tiny bit of research before. If i understand it this would help mainly with batching (ie concurrent sessions/users/turns etc) to keep the pipeline more busy (vs a batch of a fixed size going together), which is interesting, but most work so far has been on latency hiding for straight line performance to start.

We did have a version of things which used expert islands for MoE parallel alternative (I think that has been mostly scrubbed from the code). It showed early promise by having trunk+hot experts together, but as models got larger, it made both that very large on its own but also didn't seem to work as well (or we just weren't good at grouping experts), the idea being to really side step latency and route sessions to those islands. It felt like as models scaled it didn't stay smart (as diverse experts activated more than I thought they would). Could be our mistake (was an exciting possibility though - if you don't mind accepting some loss).

tldr; yes this seems a very nice enhancement for smarter batching/keeping things busy and seems like most larger models we look at are MoE!

Have been using morph for a while (I am one of the authors of goose) and was surprised when introduced at the boost it gave me (much less iteration with the main expensive LLM, and I can even make the editing process simpler to take a load off the agent). Used it with claude 3.5, 3.7, 4 and currently with a o3/openai and anthropic/claude4 + morphllm combo today.

I built this using openAI codex, google cloudrun and streamlit (visualisations by graphviz - codex is not bad at translating pipelines to graphviz without really much prompting which somehow surprised me). Let me know what you think. No fine tuning required for this (as Codex already knows quite a lot).

Agree Kube is not for everyone, no question, I was just trying to clarify what was in the article. (if you are offloading complexity - I would hope it is to a GKE or EKS or AKS or similar, in which case it is very much offloaded...)

I didn't mean to imply that you mix all those things together from the CDF - was just mentioning some interesting projects (some are unrelated) in the mix.

Agree also on simplicity - myself I don't like to run anything, so CodeShip is what we have for that (but it sounds like you are referring to self hosting only solutions?)

(edit: and thanks for insightful comment)

well as far as I know bash and gcc don't depend on kubernetes yet, but that is a good point.

It isnt' for everyone (and I wouldn't want to run it myself - GKE or EKS is fine but otherwise.. it seems tough to run).

Also this is really a starting point for Jenkins X - I expect options for non kubernetes to be coming.

Right, I feel better about doing this now!

Charles Sturt Uni is a fairly well regarded Australian university. It isn't one of the so called "sandstone club" but I believe it does legitimate stuff (it isn't that easy to be called a university here). They also often specialise in rural/earth science matters (I always thought of them as a rural uni). If that helps?

Yeah it is my instinct to to use credentials but then I pull myself up and think "is this a high brow ad-hominem thing" but it probably makes sense to be curious/skeptical at first.

>That's just the thing. Why has there been so little adoption of commonly needed functionality as part of core, or at least as officially supported plugins?

Oh there are core bundled plugins, official etc - they are just core functionality that happens to be implemented by plugins.

>And why is version control not standard in 2018?

that is and always has been - "git" support used to be not in by default, but that was a while ago (it is included now).