fswd

Right now, Qwen3 4B

I am reading this right now on my N100 desktop and it's 7W

I've seen two super nova in my life time. once in 2008 and another in 2012

using this same architecture, it would be cool to build a serverless-git

I probably have the same N100 miniforums and can only get it down to 7w. Is it possible you could share how you got it down to 3.6W and 4-8W?

I just got this N100 a few weeks ago, have it on a killawatt next to my desk.

my story with them: Chase bank refused to cash their own Chase cashiers check. Seems crazy but I had to get a lawyer involved.

yes that's the whole point of arxiv to allow anyone to publish.

> Some interesting challenges: 1. There is a React parser that is used to parse, insert the style, and serialize it back to code 2.

here's another approach I used with dynamic react

https://github.com/mhsdesign/jit-browser-tailwindcss "dynamic tailwind"

I've had the same exact experiences.

this is so bizarre? I use GraphQL with Remix in my loader. It's a great combination. 7ms response from SQL. I can't understand the article. Sorry this person doesn't understand GraphQL nor Remix. might be AI generated

The graph code, when all done and said, is identical for all scenarios.

using LangGraph for a month, every single "graph" was the same single solution. The idea is cool, but it isn't solving the right problem.... (and the problem statement shouldn't be generating buzz on twitter. sorry to be harsh).

You could borrow some ideas from DSPy (which borrows from pytorch) their Module: def forward: and chain LM objects this way. LangGraph sounds cool, but is a very fancy and limited version of basic conditional statements like switch/if, already built into languages.

like something I'd use on my amiga 500 when I was a kid!

I just finished implementing S3 file upload in nextjs to cloudflare R2 with a supabase backend. Wish I had been lazy and waited a day!

These have to be spy satellites. GPS is just a timing signal, in space. Technology we had 40 years ago. There is no way you need $100M to build this and have special government authorization to collect the debris.

don't forget the timeout reconnect!

This story is heavily censored on hacker news

https://github.com/vitoplantamura/HackerNewsRemovals?tab=rea... (scroll down to bottom)

39465301 #25 3 points 0 comments -> Google Gemini leaked its woke prompt injection process

39465255 #1 78 points 1 comments -> Google to Pause Gemini Image Generation of People After Issues

39506022 #30 -> Gemini: Who negatively impacted society more, Elon Musk or Hitler?

39508833 #17 30 points 24 comments -> Losing Trust in Google

39516032 #15 8 points 6 comments -> Gemini and Google's Culture

stuff like this makes me want to experiment with going back to just one huge $100k server and running it all on one box in a server rack.

OpenAI has broken every promise it has made

sorry I meant to say or explain computational graph, and yes langgraph was reintroduced recently last week.

Yes, I have implemented HITL. There's a task queue(psql skiplock). You make a request (insert row). It builds a "workflow graph" of your request and compiles it, a combination between a FSM and graph program. It runs an iteration on the workgraph, and then if successful, submits the next step in the queue as "pending". the user then inspects and accepts the tasks, which then run another iteration of the workgraph. If the user rejects it they can click "playground" and manually do the step.

I think they did several stock splits, and this time period was one of them

you cannot currently run mixtral with a 32k context on a 3090. Unless am I wrong? I think the largest context I was able to reproduce was around 1500 with 2 or 3 bit, I would have to look at my notes.

the management talent is preventing it

I'm so glad we're seeing memory specs next to the core count, awesome.

I solved this problem locally. When uploading a file to the server before going to S3 it is cached in redis. Whenever the codebase needs to use the file, it checks redis, and if it is not there it fetches it and caches it again.

I just bought a 10th gen i5 laptop with 16gb of ram for $60 off of ebay. HP c640 chromebook. Fast enough to run a 7B LLM at 1-2 tokens/sec

Tesla should make a fleet version that's easy to repair, and jump into the market place or help Hertz get back on its feet.

There's a partial lunar eclipse next week, October 28th https://www.timeanddate.com/eclipse/lunar/2023-october-28

warm days-old left out fermented milk called Kefir.