I love the fact that they're advertising the latest Codex Cli while most Windows users had to rollback/downgrade to an older version because of the annoying excessive terminal screen flashing issue (https://github.com/openai/codex/issues/48074)

(I see a newer version was posted an hour ago. Let's see if the issue is now fixed...)

Is this compatible with Graphify-Labs/graphify, w/r/t existing output formats?

yes it is compatible with graphify - the json can be used directly by graphify.

I'm blocked for some reason, probably too much HN referred requests...

Enable JavaScript and let the site issue a “not an AI” cookie and it should then be usable. (I have a JS button for this on my toolbar and it’s come in handy a lot recently..)

Are you serious you have a "I am not an AI" widget?

I read it as "a button for re-enabling JavaScript"

This, with state default-off and remembered per-domain. Not my invention but glad for it!

The pelican is for the last gen of LLMs -- have you tried a penguin instead?

It's actually the other way around - the evil Penguin villain from Wallace and Gromit is secretly controlling SimonW !

I'm so disappointed with the memory pricing we have to endure... MacStudio M5 Pro is $2,499 base with 36GB RAM, but when you increase it to 256GB, the price jumps to $9,499!!!

How do you use a 72GB model as your daily driver locally?

Apple Silicon. But: there's no need to use the FP16 version. At 8-bit precision the quality loss is almost imperceptible. That cuts the footprint to 36GB. Which is great for a 64GB Mac, because you have room for plenty of context. 6-bit also works nicely at 26GB + context.

You want to use the newer quantization formats like Unsloth's UD quants or oQe, where the weights are selectively quantized using a calibration dataset so that important weights are left at/closer to full precision.

At 4 bit it easily fits in 32GB. That's what most people use.

use a quantized version. since it's MoE, what matters is that the 3b parameters that are used for every token fit in gpu vram, the rest can stay in system ram. really great if you don't have unified memory.