vergessenmir

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.

DSPy is both great and frustrating at times. It's designed dort prompt and response. Textgrad and DSPy share this design decision and this is where the frustration begins.

I'd like the modules (dspy speak for inference ) to be aware of the chat history. Without this I'm not able to test the accuracy of the responses in the various chat contexts they appear.

It's been asked for in the TextGrad & Dspy GitHub issues.

it's baffling to me that realising that your application deployment requirements don't fit k8s means that k8s is not fit for purpose. Note that the alternative isn't a return to the management of VMs, which we all agree none of us ever want to return to, but is a cloud service which hides the concerns that k8s makes explicit.

It's a bit short-sighted (granted not intentionally) to talk about k8s not being fit for purpose without specifying the workloads it's not suitable for and at what scale of operations. Often times when these discussions come up it's usually reduced to a stateless web app being the justification for illustrating the shortfalls of k8s as a solution.

Personally, I'm not a big fan of k8s because it can be complex but the alternatives tend to be more complex. I can guarantee, your solution over time will converge on operational knowledge trapped in the minds of your team members , runbooks and custom scripts that express an abstraction that k8s implements perfectly well.

is there a plugin to convert the ast to json? Similar to the mistune package in python. I'm using this as part of a rag ingestion pipeline and working with markdown ast provides a flatter structure than raw html

The repo has sample code and it is fairly easy to create a node that will do it.

You won't however have access to usual sampler, latent image, Lora nodes to do anything beyond basic t2i

To give you a bit of perspective,the democrats are right of the Conservatives in the UK.

So they would kinda feel feel far-rightish to us only because the democrats are more conservative than ours

VC firms are betting on the people. Most early startups are still looking for their market fit. The VC firm is betting that the people are able to identify that market which they'll be able to scale.

What is the "think step by step" prompt? An example would be great, Is this part of the system prompt?

My grandfather died over a hundred, he had 16 children and two wives. We estimated his age to be at 103 going by the youngest possible age he could have become a father.

My grandmother is well into her 90s.

They both were active throughout their lives, always in the fields.y grandmother goes to bed when the soon after the sun goes down. She insists on not having any electricity.

Makes me feel old. I was at uni when Ubuntu came out. But my story was similar to yours in the mid 90s and I got hold of a walnut creek cd with Slackware from some PC mag, a couple floppy disks and discarded hardware and I was off to the races

There was something about discovering tech through dedicated tech sites back then that felt exciting.

Now, any time I find something new it always has a polished marketed feel to it and has none of the secretive clandestine undiscovered power that old tech had.

I guess I am getting old

I use zen. Ditched firefox and brave irrationally last week because youtube runs so much better and page loading is fast. Let's see how long this change lasts

I'm sorry,but I can't tell if you're being serious or not since you commented without qualification.

One of the most stable system archictectures I've built was on Kafka AND it was running with microservices managed by teams across multiple geographies and time zones. It was one of the most reliable systems in the bank. There are situations where it isn't appropriate, which can be said for most tech e.g( K8S vs ECS vs Nomad vs bare metal)

Every system has failure characteristics. Kafka's is defined as Consistent and Available and your system architecture needs to take that into consideration. Also the transactionality of tasks across multiple services and process boundaries is important.

Let's not pretend that kubernetes (or the tech of the day) is at fault while completely ignoring the complex architectural considerations that are being juggled

No screenshots, no example customisations. The installation is not exactly off to a great start either:

(venv) user@kubuntu:/mnt/sdc1/github.com/glacier$ pip3 install -r requirements.txt

Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com Collecting PyQt5>=5.6 Downloading PyQt5-5.15.11-cp38-abi3-manylinux_2_17_x86_64.whl (8.2 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 8.2/8.2 MB 1.6 MB/s eta 0:00:00 Collecting PyQt5-sip<13,>=12.15 Downloading PyQt5_sip-12.15.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.whl (270 kB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 270.5/270.5 KB 3.0 MB/s eta 0:00:00 Collecting PyQt5-Qt5<5.16.0,>=5.15.2 Downloading PyQt5_Qt5-5.15.14-py3-none-manylinux2014_x86_64.whl (60.5 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 60.5/60.5 MB 2.1 MB/s eta 0:00:00 Installing collected packages: PyQt5-Qt5, PyQt5-sip, PyQt5 Successfully installed PyQt5-5.15.11 PyQt5-Qt5-5.15.14 PyQt5-sip-12.15.0

(venv) user@kubuntu:/mnt/sdc1/github.com/glacier$ python main.pyw

I got 0 out of 10, turns out they wanted me to select the correct photo and not select the fake. How was this obvious to everyone else, certainly wasn't to me

The nice thing about obsidian is that you can start off really simple. It can be as minimal as linked markdowns documents if you like. Which is pretty much most of the core plugins disabled up to all the bells whistles.

I've never used org-mode so can't really comment on whether you'll actually be gaining anything

Waitrose tills and card machines aren't working

Jetbrains has a decent golang IDE. I don't use it because I use Vim. Vim's not for everyone.

It will allow for more realistic emotions in current SD Model merges and fine tunes by generating frames correctly labelled with their associated emotions.

Most SD1.x/SDXL models images depict humans with the same expression so the frames generated by LivePortrait will help with training datasets.

I believe the Pixar animators in Toy Story 1 used facial expressions /emotions database called F.A.C.S to make the characters more humanly relatable.

It's not clear if the "expressions" will generalise to new faces

Showing my age here but povray is what got me into software engineering. I wrote a raytracer in Pascal + Assembly and then in C and ASM

This was in 1995.

Why do ML, especially nowadays on a Mac, when you can do it on an Ubuntu based machine.

Surely work can provide that?

Isn't this more of A semantic understanding of how language maps onto geometry rather than understanding of actual space and geometry?

I sometimes feel like this may not work without language helping it along

A 100% agree. I've built so many of these solutions over the years and haven't quite cracked the culture problem

I think your definition of observability sits on top of the other two (Control & Cognitive definitions) and ultimately is more of a description of an approach to observability rather than a redefinition of it.

Observability is hard and I agree with the approach you've laid out, especially on the involvement of leadership insight. It's crucial and it has been omitted or ignored while working on this for a decade. The tooling improves but the same organisational problems persist. And there's an over reliance on hyper-competent people to get this working well than on an agreed set of practices that help offset the average competency that exists in most places. Don't get me wrong we need competent people but I find that when they leave so do the good practices.

Redefining it is not going to do much since these are organisational, cultural problems which all happen to be the issues that hobbled DevOps and the SRE movements and we all know how those movements turned out - they never matured into any of the promises they made.

I'm not sure that this is a Docker problem but you do have a point. I've used docker from the very beginning and it always surprised me that users opted to use package managers over downloading the dependencies and then using ADD in the docker file.

Using this approach you get something reproducible. Using apt-get in a docker file is an antipattern

This project is leaving a bit of a sour taste in my mouth, the more I read through it. It may just be because it is early days and I hope the criticisms/questions are taken constructively. Here are question to the OP:

- How are multiple environments managed, is my hard disk going to fill from multiple overlapping virtualenvs - How are resources managed for parallel runs, or is this offloaded to the cloud? - It lists Mac OS and Windows as supported platforms, are there plans for there to be a Linux version? (It is a bit surprising the Linux is omitted)

The non standard licensing puts me off in contributing or using this. It is frustrating how the phrase opensource has been diluted in the AI/ML community. ComfyUI has a GPL license [1] while this project uses this [2]. I honestly don't know where I stand since this is a legal document using non-standard phrasing to describe how the rights around the source code.

This is a project that uses a custom license with less rights provided than the ComfyUI project it self-describes as improving. Am not sure the title is reflective of the project.

[1] - https://github.com/comfyanonymous/ComfyUI/blob/master/LICENS...

[2] - https://github.com/6174/comflowyspace/blob/main/LICENSE

I would imagine with an img2img workflow it would be. The same way you can reconstruct a badly rendered face but doing a second pass on the affected region

Reasoning across many stages, converging on a user provided goal with the required level of accuracy is beyond commercially available LLMs. Take the travel agent use case, a recent paper showed that Llms tested would get dates and prices wrong. So the promise of AutoGPTs, GodGPTs etc is still quite far away

I'll take prompt adherence over quality any day. The machinery otherwise isn't worth it i.e the controlnets, openpose, depthmaps just to force a particular look or to achieve depth. Th solution becomes bespoke for each generation.

Had a test of it and my option is it's an improvement when it comes to following prompts and I do find the images more visually appealing.