It's 3am and we're arguing some insipid minutae over technically illegal tequila shots while one drunk girl is breaking it down on the tiny dance floor :)
What does it do when the model wants to return something else, and what's better/worse about doing it in llamafile vs whatever wrapper that's calling it? How do I set retries? What if I want JSON and a range instead?
You can't do it as part of whatever's calling it because this changes the sampler. The grammar constraints what tokens the sampler is allowed to consider, only passing tokens that are valid by the grammar.
I've had a bunch of windows replaced over time. It is expensive but some had gotten really ugly looking--in addition to being, I'm sure very inefficient. I have oil heat and have just never seen the justification for solar especially given my electrical usage and all the scammy solar stuff out there.
FWIW, https://indowwindows.com/ (San Francisco) make window inserts that are the exact same size as the existing window, plus a wooden frame, so they look exactly like the existing windows. Mostly for keeping the exterior look of buildings in SF due to regulations, they block out sound and help with temperature control. They're pricey tho.
As for solar. If you have the budget and the disaster preparedness for it, there are some camping grade solar + battery systems that aren't house-grade, but great for power outages. The systems range from small (affordable), to the 1000 Plus, consisting of a battery with 1264Wh capacity and 120 v inverter, along with 200W of solar panels. That one'll run. you $1,500.
Do you have an Indow? I’ve been eyeballing them for a while, but they’re pricy enough I haven’t been willing to call the shot yet.
I do have a pretty big bluetti and a couple panels for it - I also grabbed an RV fridge for storing food. Naturally, the capitalist gods smiled upon my financial offering and have blessed me with steady power since then.
> With a maximum take-off weight of about 16 tons, a 25-meter wingspan and an operational ceiling of 15,000 meters, Jiutian can remain airborne for roughly 12 hours and reach targets up to 7,000 kilometers away
Okay, you're right... their military capabilities are still laughable. Thanks for the reassurance, that'll help me sleep at night a bit better.
paying to create the nodes isn't the hard part. The hard part is the protocol isn't designed for this and is easily overwhelmed. In case of adversarial attack, it's trivially going down because the protocol is easily attacked, as we see at any popular gathering like Defcon.
How hard would it be to use with OpenAI's offerings instead? Particularly, imo, OpenAI's better at "looking" at pictures than Claude.
reply