Jam321

Live free or die.

Open ledger minimalist, paranoid crypto anarchist.

[Jam321 @cryptoanon69]

To all small block bitcoin maxis and transparency cucks..

[ᴜɴᴛʀᴀᴄᴇᴀʙʟᴇ @DontTraceMeBruh]

The future is lunarpunk

[Jam321 @cryptoanon69]

@rohanpaul_ai Now they just need to make something that takes proper advantage of ssd offloading

[Jam321 @cryptoanon69]

@prayag_sonar Torrent site for llms is inevitable. Where is the top alternative today @grok ?

[Jam321 @cryptoanon69]

@Tono_Ken3 You have way more compute this way. Compute is what matters, nothing else.

[Jam321 @cryptoanon69]

@Tono_Ken3 I wonder if just another model would pick it up as well, like qwen flash, not necessarily a frontier one

[Jam321 @cryptoanon69]

@morganlinton @loktar00 Get a 3090, keep the 20$ sub. Split your personal data and coding projects on two machines.

3090 + qwen and your custom setup for being your personal assistant in everything, give it all your data.

20$ frontier for coding stuff.

Buy a 3090 just to code? Only if complimentary

[Jam321 @cryptoanon69]

@toraxx @ivanfioravanti Exactly, so maybe run the 27b properly at decent speeds instead was my thought. A lobotomized 125b is fun i guess, but why

[Jam321 @cryptoanon69]

@0xSero They have 27b, it's kind of the best model ever made

[Jam321 @cryptoanon69]

@MartiniGuyYT @coinbureau It isn't x, its y

It's not just y, its xy

Obvious ai slop poster

[Jam321 @cryptoanon69]

@AskKlarna @0xBOYD @Javi @Klarna ??😂

[Jam321 @cryptoanon69]

@MikeBradleyAI @TechMDAI Qwen flash is gonna be🔥🔥 on those bruh.

[Jam321 @cryptoanon69]

@loktar00 The rtx will be another world when it comes to live interaction. It would enable proper vision, hearing, speech. Look at the qwen3.8flash. its not long until you have a model like that with everything fast inside 96gb vram and huge knowledge bank on ssd. Proper jarvis tier

[Jam321 @cryptoanon69]

@MickaelV79228 It came out now

[Jam321 @cryptoanon69]

@Alibaba_Qwen Finally, my dream might come true with Qwen4🙏

Any reason why not a much much larger N-gram wouldn't help?

[Jam321 @cryptoanon69]

@Alibaba_Qwen Finally, my dream might come true with Qwen4🙏

Any reason why not a much much larger N-gram wouldn't help?

[Jam321 @cryptoanon69]

@QwenDevs How about making a huge 500-1T+ model with less than 10b active parameters? Or even a3b?

Why can't we have a 27b dense model that can quickly find everything it needs through an ultra sparse model cached in ram + ssd? A tiered model structure.

[Jam321 @cryptoanon69]

@GrapheneOS @MurenaCom Wild

[Jam321 @cryptoanon69]

@thedave9081 @ArmitageX @AncapAir Fuck

[Jam321 @cryptoanon69]

@tomgreenwald The halo's are much cheaper vs those. Are those prices really reflecting reality? Spark has gone up to above 6k here, while halo is at 4 still