vergessenmir

it's fast for the classes

I am building a tool to manage your image generation inference. It uses ComfyUI as the backend and it allows you to index, search and invoke your workflows.

It makes it super easy to using existing workflows to chain them together into more complex outputs.

All of this withouts nodes.

Early release is out here: https://github.com/svenhimmelvarg/kaleidoscope

I agree with you but the main thing g here, IMO, is the friction with all the alternatives you mention in getting something working.

For example, I've never heard of Automator. I'm familiar with Zapier, I'll have to evaluate the two situations, then I'll find out that might need to find an alternative that runs on Linux and then I'll have to check if....

These are all simple steps but they all use a non-trivial amount of time for the problem their solving

The other thing is the

Just pay for the sync. I used to juggle with git, rsync, inotify etc and other tools

Its one of the few subscriptions where it actually feels like money well spent

If you don't have the experience you can't provide it with stylistic guidance, or idiomatic patterns or provide examples to direct it.

This leads to the idea that LLMs with existing languages can't really learn new idiomatic patterns.

For new engineers I think new paradigms will emerge that invalidate the need to know the current set of design patterns and idioms. Look at the resurgence of unit tests or the new interests in verification systems.

Package situation on anything that isn't Arch (and I think Fedora) is pretty rough. I installed it from source. It helps that it is a rust application and was up and running in no time

See my comment above (moved from i3wm) but my spec is

RTX 3090, Pop OS 24.04 (beta), 4K 43" Monitor,

Nvidia cards worked out the box with no problems

Moved onto Niri yesterday after having to reinstall my PopOS and it just clicked. Like i3wm did all those years ago.

I can focus for hours on end and spend zero mental energy on resizing a window. I had less of that with i3wm but you had to always readjust after a few windows were tiled to your workspace. That final bit of cognitive overload was removed with Niri.

EDIT: Spec: RTX 3090, Pop OS 24.04 (beta), 4K 43" Monitor,

Niri Installed from cargo build, super easy install, make sure you install xwayland-satellite so that you can run VS Code, Obsidian, Zoom, Blender and other strictly X11 applications

Harbor has its pain points but it is infinitely easier to get up and running compared to crufty Artifactory.

One glaring omission is lack of support for proxy docker.io without the project name i.e pulling nginx:latest instead of /myproject/nginx/nginx:latest

The workaround involves URL rewrite magic in your proxy of choice

I don't get it, what does this do?

Is there any notable properties of this implementation, are some parts slower, faster etc

get work done, look younger and slice off the first 5 years of your experience because it is "not relevant". I look about 10-12 years younger so I am able to slip under the radar but it makes me wonder how my peers who look visibly their age fare? The job market is london, sector: Hedge funds, asset mangement tech etc.

Go is great for concurrency. Not quite there for agent support. The problem isn't performance or message passing it's the agent middleware i.e logging, tracing, retries, configuration

You need a DSL either supported in the language or through configuration. These are features you get for free in python and secondly JavaScript. You have to write most of this yourself in go

some writes might fail, you may need to retry, the data store may be temporarily available etc.

There may be many things that go wrong and how you handle this depends on your data guarantees and consistency requirements.

If you're not queuing what are you doing when a write fails, throwing away the data?

I'd probably check this out in my home lab but as a corporate user the offering of discord as a support channel makes me nervous.

Discord is predominately blocked on a corporate networks. Artifactory ( & Nexus) are very common in corporate environments. Corporate proxies are even more common. This is why I'd hesitate. These are common use cases (albeit corporate) that may not be readily available in the docs.

I'm not sure what the rush is. Do we have to make it profitable now? 2 years is not a long time. The point has been made elsewhere by other commentators about the dot-com bubble and how long that took for trillion dollar industries to form. It sounds like his gripe with Altman's hype narrative has soley informed his somewhat negative view on LLMs as a whole.

I find it interesting that he almost equates OpenAI === LLMs and misses the fact that the hype is not purely industry driven. For instance, the number of machine learning papers in the last year has quite literally doubled.

This is also typical of an Americentric view on innovation that we don't report on the quiet revolution happening in education in underdeveloped countries that are a direct result of the accessibility of this unprofitable technology.

I don't think we need killer application right now

We also forget the internet bubble happend first

I think the author is looking at LLMs through the lens of Sam Altman's hype narrative and I wonder why we care so much that

I feel this particularly when I use gguf support.

How do you get accurate information on the template structure?

Could you expand on this, is there any disadvantage to continuing with ollama?

I use Ollama for prototyping and then move what I can to a vLLM set up

It's all kinda hidden. The documentation steers you towards SSR or assumes it. It appears as though you can't have client side routing without a roundtrip to the server. Take a looked at the "page" documentation immediately under Routing.

How do I serve svelte files using a python or golang backend and still have client side routing? These should have a fairly straightforward answer but I don't think they do.

The push of to SSR in Svelte5 has been so off putting. I know CSR is possible but 1) it's not easy, and 2.) practically undocumented

How do you get access to 40M Hotel Photos?

Hah, more license confusion! It is a good learning resource for those wanting to see how such a codebase works. I do think they should have opened the repository as read-only and that may align more with the educational intent.

"We will only consider pull requests and issues regarding self-hosting or critical fixes. "

Having this makes the messaging a bit confusing. You will accept improvements and bug fixes which we can use commercially but you can't - Did I read that right?

I used to be heavy Anki user but I find the setup a little hard to follow.

The image generation can do with ComfyUI integration. This will expand the illustration options beyond stable diffusion. It will also allow for more options within stable diffusion too.

I might raise a PR tomorrow if I have the time.

Is any of this reproducible? Not sure why that requirement has been quietly overlooked.

I've worked in this space for a long time and can't make head or tail of what glue is.

A motivating examplt would be help which I might have missed?

It's why I prefer Claude to let's Chat GPT. I'm paying for tokens generated too.

Tech debt is a function of your code base, it's age, team turnover and number of pivots. Many factors to consider but I'll focus on pivots.

You can't plan for a pivot because it's a known unknown. The same way you can't plan for a specific financial event in the market but you can brace yourself for a category of scenarios. Even with that, you can't predict the impact or the appropriate response your business needs to take.

In the same way so is the pivot. The nature of the pivot is the market revealing the debt you didn't know you had. The magnitude of that readjustment to the market, in the time it has to happen and the time to the next pivot is unknowable because it's information not present at design time.

When working with GGUF what chat templates do you use? Pretty much every gguf I've imported into ollama has given me garbage response. Converting the tokenizer json has yielded mixed results.

For example how do you handle the phi-4 models gguf chat template?

Wouldn't try and convince the nay sayers. I have a few comments (or opinions rather) This coming from a recent dspy user hitting its pain points.

- The GitHub page is very busy

- A clear example should come up early on the page. It's only when I got to the fiddle I could see a motivating example i.e the extractions, functions and tests.

- Then a section for running tests/evaluations

-Then deployment or run with/without the baml cli

- I do wonder if all the functions have to be so tightly coupled with the model. In dspy my modules are model agnostic and I can evaluate behaviour across different models. It's not so clear how to do this

And have the chat history end up as an input field instead of each request response being prepared individually for the LLM endpoint?

I'm new to the library but from what I can see the Chat adapter will do this automatically if I use the forward call

Quoting the docs: "Though rarely needed, you can write custom LMs by inheriting from dspy.BaseLM. Another advanced layer in the DSPy ecosystem is that of adapters, which sit between DSPy signatures and LMs. A future version of this guide will discuss these advanced features, though you likely don't need them."

I could be wrong, I could be looking for complexity where there is none.

Have a fundamentally misunderstood how this all works, it sometimes feels like I have?

Wouldn't try and convince the nay sayers. I have a few comments (or opinions rather) This coming from a recent dspy user hitting its pain points.

- The GitHub page is very busy

- A clear example should come up early on the page. It's only when I got to the fiddle I could see a motivating example i.e the extractions, functions and tests.

- Then a section for running tests/evaluations

-Then deployment or run with/without the baml cli

- I do wonder if all the functions have to be so tightly coupled with the model. In dspy my modules are model agnostic and I can evaluate behaviour across different models.