Dan Southwood-Wells

I work at @EskomSePush ⚡

[vLLM @vllm_project]

vLLM v0.22.0 is out! 459 commits from 230 contributors (63 new). 🎉

Highlights: DeepSeek V4 hardening (NVFP4 fused MoE, full+piecewise CUDA graph, ROCm support), experimental Rust frontend in-Tree, batch-invariant Cutlass FP8 (28.9% lower e2e latency), Model Runner V2 advances, multi-tier KV cache offloading.

Thread 👇

[Georgi Gerganov @ggerganov]

llama.cpp now has an official website: https://t.co/vztdUpdBWL

Our goal is to make local AI accessible to everyone, and improving the user experience is a big part of that. On the new landing page you’ll find a single-line cross-platform installer. The installation provides a single unified `llama` entrypoint which you can use to run/serve models and interface with 3rd-party agentic applications.

While oriented towards simplified user experience, the new `llama` application also provides all the advanced functionality of the existing llama.cpp tooling with which experienced users are already familiar. Also note that all GGUF models that you might have already downloaded with llama.cpp in the past will be automatically available to use without downloading again (they are stored in the common HF cache on your machine).

We have many improvements in the pipeline both at the UX and at the engine level and we plan to iteratively ship new things over the coming months. One of the main focuses will be seamless integration with local-friendly 3rd-party agents (such as Pi). In the meantime, we’ll continue to listen for feedback from the community and adjust accordingly, so keep letting us know what you think and need.

[Dan Southwood-Wells @c4talyst]

Amount of time I spent at my desk this week: patties

Configured a Frigate object classification for desk presence, exported from HA to ChatGPT and asked to give me fun graphs to show how useless my work week was 😅

Had some fun Friday deploys though 🚀

[Dan Southwood-Wells @c4talyst]

Being sick (sinus) is like coding with a 9B parameter model, crazy times ahead, good luck

[Dan Southwood-Wells @c4talyst]

This is great and stuff, but what's happening to all that perfectly good wide open land in what was District 6 ? 👀

Why does no one talk about it? Make it right.

[Dan Southwood-Wells @c4talyst]

Discovery Health being shady here? Get points, you just need to share your 'Electronic Health Record' ...

🚩This means sharing everything you discussed in the room, not just what you've been treated for (ICD-10 codes)

[Dan Southwood-Wells @c4talyst]

@hermux US rebounding

[EskomSePush (ESP) @EskomSePush]

1 year with no loadshedding 🎉
Since 15 May 2025 at 22:00. Congrats.

[Herman @hermux]

Can you please start naming our storms @SAWeatherServic ?

I think it will make the news stories more fun.

"Cassiem blew off my roof"
"Gareth floods Hermanus Mall"
"Mandisa sweeps through Western Cape"

I'm ready to help

[Herman @hermux]

Oh and you can monetise by allowing sponsored names 💰

"Sixty60 arrives unexpectedly fast"
"Pronutro ruins everyone's morning"
"Households destroyed by Hollywoodbets"

[Dan Southwood-Wells @c4talyst]

It's a bit awkward that I'm now getting emails from the DA, on an email address I only ever gave to DSTV

[Dan Southwood-Wells @c4talyst]

@JustinPBarnett Was playing with qwen3.6-35b-a3b earlier - quite snappy in comparison to 27b

But I need to optimize

[Dan Southwood-Wells @c4talyst]

@ruark @JustinPBarnett Not doing coding with this machine - yet.

Doing text & image stuff - summaries, moderation, intent, etc.

[Dan Southwood-Wells @c4talyst]

@hermux At the similar time other big players (i.e. X and Reddit) started closing their public APIs and blocking scrapers ; meanwhile stack overflow ... Sold 🤣

[Dan Southwood-Wells @c4talyst]

@ruark @JustinPBarnett We run an anonymous chat app 😅 Lots of stuff, it's in production and removed most of our hosted model costs (besides dev).

[Claude @claudeai]

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity.

This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

[Dan Southwood-Wells @c4talyst]

Power outage Orlando East

[Dan Southwood-Wells @c4talyst]

@wootware Here we go 🚀

https://t.co/fvwFltUiOU

[Dan Southwood-Wells @c4talyst]

The homeserver got a dual 3090 upgrade. Now running Qwen3.6-27B-FP8

[Dan Southwood-Wells @c4talyst]

Thanks @wootware for sourcing me a PCIE PSU cable - I needed 5 and only had 4 😅 You peeps are heroes 🚀

[Dan Southwood-Wells @c4talyst]

@chriszeuch Excited for the one week later review 😅

I hate it, so much 🤣 I think it's a polarizing device