Harrison Kinsley

gpus and tractors.

Director of AI and Engineering @ networks from Scratch book:

[Harrison Kinsley @Sentdex]

Learn to program neural networks from scratch in Python.

Forward pass, back propagation/optimization, and deployment, all line by line, w/o 3rd party libraries.

[Harrison Kinsley @Sentdex]

@andymstone meta pushes software to smart glasses for features that dont exist?

[Harrison Kinsley @Sentdex]

@tbpn Bro lol

[Harrison Kinsley @Sentdex]

@jonstephens85 @NVIDIAAIDev Did u get to ride in one?

[Harrison Kinsley @Sentdex]

@TheAhmadOsman @Tnimbus @NVIDIAAI i take it back. made the attempt lol

[Harrison Kinsley @Sentdex]

@TheAhmadOsman @Tnimbus @NVIDIAAI p sure it can run on 3x

[Harrison Kinsley @Sentdex]

Free endpoint goes brrrr. Apparently I'm putting hermes on the map for nemotron ultra

[Harrison Kinsley @Sentdex]

Everyone stop what you're doing right now and use this free openrouter endpoint with hermes agent: https://t.co/oWuE8x2AOH

and tell me how absolutely terrible it is so I am not pressured to purchase more GPUs.

thank you for your help in this tough time for me.

[Harrison Kinsley @Sentdex]

@fares_lass you would be wise to not believe any api that tells you your data isn't being logged :)

[Harrison Kinsley @Sentdex]

@sinecanswork lots of models are free for a few days or weeks when they first release. Always a good idea to check openrouter and just search "free." Almost always some free 100B+ parm models.

The only stipulation really is like... 99.99% chance they're logging everything and using that data.

[Harrison Kinsley @Sentdex]

@ammmanism sounds like me!

[Harrison Kinsley @Sentdex]

@austimullins afaik, it does. I do seem to recall there were hiccups with the sparks and maybe rtx pro 6Ks initially, but I think that's long resolved. Hope I'm not mistaken xD

[Harrison Kinsley @Sentdex]

That's a big boy. Gonna need a few more RTX Pro 6Ks for this one.

[NVIDIA AI @NVIDIAAI]

@openclaw @NousResearch @LangChain As always, Nemotron 3 Ultra is fully open.

This includes model weights, synthetic data, and post-training recipes. Available now on @huggingface → https://t.co/MDdnY047fw

[Harrison Kinsley @Sentdex]

@SearchDataEng pls help

[Harrison Kinsley @Sentdex]

@MaxCalkin Oh, right lmfao. Duh! Thanks :D

Dang I'm going to be the #1 token user for this entire model shortly lol. That's crazy!

[Harrison Kinsley @Sentdex]

@MaxCalkin whered you see only 13M tokens I wonder? I tried to see this stat and it looks like 652B tokens

[Harrison Kinsley @Sentdex]

@edwixxxx this has got to be a bug. I refuse to believe this is a thing with this model lol.

[Harrison Kinsley @Sentdex]

@MaxCalkin aaaaaahahaha, mate that is actually mostly me.

[Harrison Kinsley @Sentdex]

@InnoboSJ is it? I've not seen a modern model do this in a long time. I would find this shocking if this is an expected or remotely common behavior!

[Harrison Kinsley @Sentdex]

@edwixxxx I went to openrouter to see how much it'd cost me to try it, and it's free there rn if you want to give it a spin: https://t.co/oWuE8x2AOH

[Harrison Kinsley @Sentdex]

@jhamann93 My calc is you need apx 335GB minimum, 400+ ideally. I think 4 RTX Pro 6Ks will do it @ NVFP4 but there will be likely at least a couple 4bit quants that will fit on 3 RTX Pro 6Ks. We'll see. In @UnslothAI we trust.