Thank you, I appreciate your response! So it is the primary way to benchmark how the klankers are doing. For I am rich, life is good. I have family, friends, my health and a roof over my head…

I see, one of those are like 40% of my annual income, not a chance for me to buy even one T-T

Hmm not sure I have the hardware for that

FP8

What quant?

Damn, doesn't that requires absurd levels of hardware?

wow, thats a nice setup

Why are you counting tokens? The tokens don’t matter since it’s all local right?

which hardware do you got?

What version of DeepSeek are you running?

I've never been able to trust qwen 27b for dev ops. As long as it has good memories and skills to make sure it plays nice. Sometimes though it gets off track and does amazingly stupid things in my experience.

Interesting setup! Consider token efficiency and model specialization to optimize performance.

So you work in HR now?

100% Artisanal Yo😀

I just installed murmur to help me do the same 😂🤣

What UI is that?

Noice. I'm building something similar here with Hermes: https://gitworkshop.dev/[email protected]/hermes-nostr-gateway

What setup is this? Hermes?

Running all of it on your own hardware is the part that matters, and memory bandwidth becomes the ceiling fast. A consumer card holds about a tenth of the bandwidth of a datacenter part, so every agent that needs its own context has to take turns. My own setup went the other way: fewer agents with n...

So cool 🔥👏🏻

Which dashboard to you use to manage them? Is that Slack, Matrix, Discord?