andyg5000

Drupal Canvas is here! Here’s how to get caught up: 🔹Read the release announcement on @Dries’s blog: dri.es/drupal-canvas-… 🔹Join the @DripyardStyle webinar next Thursday for a deep dive: dripyard.com/blog/dripyard-…

@UnslothAI any plans to roll a linux/arm64/v8 compatible version of studio for docker? I'm trying to run it on a gx10/dgx spark.

@Teknium @glitch_ @sudoingX Looks like that actually fixed the issue. Makes sense, thanks!

@AndyG5000 @glitch_ @sudoingX I think most people with qwen disable thinking because it loops but I'd wait for his expertise!

hermes agent and qwen 27b feels unmatched

@Teknium @glitch_ Using Qwen3.5-27B-Q4_K_M.gguf with a 3090 attempting to follow what @sudoingX is doing.

@glitch_ @AndyG5000 Depends on model maybe?

@sudoingX Tried this today and had a lot of issues using qwen 3.5 (3090) and nemotron (dgx) Biggest problem was loops of responses when asking to create simple reminders. I'm just using local llm option, but something is missing

I will help anyone migrate from openclaw slop bloat to Hermes agent, the majestic agent, the only one you will ever need. You deserve better tools in this changing landscape. Let's do an upgrade to your setup, anon.

@nmdmatt Congrats on this! I recommend this for anyone doing web development, not just Drupal. It covers a lot of techniques within Drupal, but also beyond the PHP request/response cycle. It's insane how powerful the Drupal caching stack is and this is a DEEEEP dive 😀

My book Understanding #Drupal: A Complete Guide to Caching Layers is now published on LeanPub! Covers the full stack — render pipeline, cache backends, HTTP headers, reverse proxies, and writing cacheable code. mglaman.dev/blog/understan…

@steipete @karpathy @nummanali You can use tmux in the shells that get launched, but I made it tmux "first". Maybe I missed something?

@AndyG5000 @karpathy @nummanali that has native tmux support, what did yoi change?

Claude Code teams with tmux is really cool When you run with team mode enabled in tmux, it automatically opens the additional terminal in pane I don't really get my main agent to orchestrate, I chat to them myself CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=true claude

@gregisenberg x.com/AndyG5000/stat… 🤷

karpathy just broke the internet with something called auto research it’s basically an ai research agent that runs experiments for you 24/7 you give it a goal like “make this model better” “find a higher converting landing page” “lower customer acquisition cost” then it runs a loop: 1) plan an experiment 2) edit the code or config 3) run a short test on a gpu 4) read the metrics 5) keep the winner 6) try again over and over while you sleep by the morning you wake up to the best version actual tested improvements think of it like a robot research intern that runs hundreds of experiments and only keeps the winners this is link to his repo github.com/karpathy/autor… for your to mess around with it in the latest episode of @startupideaspod i break down: • what auto research actually is • how it works step by step • 10 business ideas you can build with it • how to install it and start using it this one is saucy because tools like this change how startups get built watch

@karpathy @nummanali I need this too. I Started by forking @steipete's vibetunnel for tmux support (github.com/andyg5000/Vibe…). Supports attaching to multiple tmux sessions via a browser. Still a work in progress though.

@nummanali tmux grids are awesome, but i feel a need to have a proper "agent command center" IDE for teams of them, which I could maximize per monitor. E.g. I want to see/hide toggle them, see if any are idle, pop open related tools (e.g. terminal), stats (usage), etc.

How many people actually know what @karpathy's autoresearch actually does vs just vibing off the name and creating click-bait for the next hot business idea?

@harshil1712 Check out @sudoingX posts on running llama cpp flags (mostly around nvidia 3090 GPU) and @UnslothAI docs on running harnesses like Claude code locally.

Last night I tried deploying a model to a GPU. The model I was planning to deploy is 120 GB+ and of course my mac can't handle it. Looking for a smaller model now

@sudoingX Just got a 3090 last week so your timing with all this testing and posts has been fantastic. Thank you!

if you have a single RTX 3090 and want the best local inference setup right now, here's what i landed on after testing 5 open source models across 7 GPU configs this month. GPU: 1x RTX 3090 24GB model: Qwen 3.5 27B Dense Q4_K_M (16.7GB) context: 262K (native max) speed: 35 tok/s generation, flat from 4K to 300K+ reasoning: built in chain of thought, survives Q4 quant config: llama-server -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 what this gives you: - 27B params all active every token - no speed degradation as context fills - full reasoning mode on a consumer GPU - 7GB VRAM headroom after model load tested MoE (faster but less depth per token) and dense hermes (same speed, degraded under load). qwen dense hit the sweet spot for single GPU. more architecture comparisons dropping soon. what's your single GPU setup? curious what configs people are running.

@sudoingX Realized 124 tok/s was at 4K context, not 262K. vLLM eats most of the 24GB leaving almost nothing for KV cache. Apologies for the goose chase! --max-model-len 4096 --gpu-memory-utilization 0.92 --max-num-seqs 2 --kv-cache-dtype fp8 --enable-prefix-caching --language-model-only

@sudoingX replicated on vLLM. RTX 3090, Qwen3.5-35B-A3B AWQ 4-bit. 124 tok/s. key flags: --kv-cache-dtype fp8, --enable-prefix-caching, --no-enforce-eager (CUDA graphs), --language-model-only Beating the 112 tok/s llama.cpp number with batched serving ready out of the box!

drupalorg-cli 0.8.0: native GitLab issue fork and MR commands, agent skills for AI-assisted Drupal.org contributions, and an MCP server. mglaman.dev/blog/drupalorg…

If you're starting a #Drupal site-build, you'd be nuts if you didn't at least consider building your theme from one of the beautifully engineered themes from @dripyardstyle 🎨🏗️ dripyard.com/blog/meridian-… @mikeherchel introduces "Meridian". ✨🚀

Carlos Ospina @camoa proposes a blueprint for sub-$2,000 Drupal projects in emerging markets. Leveraging Drupal CMS 2.0, Recipes, Canvas, and supervised AI workflows, he argues standardisation can reduce build time without losing scalability. Details: bit.ly/4cOnxzX

Carlos Ospina @camoa releases structured Claude-based tools for Drupal devs and designers. New features include plugin architecture support, a brand content generator, and QA auditing via DDEV. Explore his open-source Claude Skills repo: bit.ly/3YAYSXb

ICYMI : The @DripyardStyle Drupal Canvas webinar recording is up. There are also some coupon codes in there if you're interested in landing a sweet Dripyard theme before years end! dripyard.com/blog/dripyards…