Steve @dbstev

@TheAhmadOsman I have one rtx 3090 + 64gb DDR4 and pretty happy. Qwen 3.6 27b fully on vram and 35b with MoEs offloaded to RAM. ~30tps on 27b and ~50tps on 35b. Also using TheTom llama.cpp exp turboquant branch, so 4x55k kv cache slots.

0 0 1 95 0

View Details

Steve @dbstev

a month ago

@AlfieJCarter 31

0 0 0 3 0

View Details

Steve @dbstev

a month ago

@mischavdburg Amazing, very good!

0 0 0 9 0

View Details

Steve @dbstev

2 months ago

@ThePrimeagen L O L

0 0 0 2 0

View Details

Steve @dbstev

2 months ago

@ThePrimeagen 🤦‍♂️

0 0 0 1 0

View Details

Steve @dbstev

2 months ago

@AFpost Yeah... 🤦‍♂️

0 0 0 3 0

View Details

Steve @dbstev

2 months ago

@dhh I couldn't agree more! It's always good to use critical thinking to challenge the status quo. Let's build what _really needs_ to be built!

0 0 0 13 0

View Details

Steve @dbstev

2 months ago

@stevibe @pupposandro Amazing model! I'm getting ~35tps on my rtx 3090 with Q4_K_M from Unsloth on llama.cpp (experimental turboquant branch) and 256k kv cache np4 Happy days

0 0 2 47 0

View Details

Steve @dbstev

2 months ago

@theo @FrameworkPuter This is great! I really liked my logitech keyboards with touchpad - but they are dated! Looks like this is a good upgrade!!

0 0 0 8 0

View Details

Steve @dbstev

3 months ago

@dhh I never thought I would go back to Intel in this life (M series + AMD here). But I'm tempted now! Gonna get a XPS 14 to play around

0 0 2 33 0

View Details

Steve @dbstev

3 months ago

@theOpusLABS @karpathy Looking great!

0 0 1 44 0

View Details

Steve @dbstev

3 months ago

@NetworkChuck Yes!

0 0 0 7 0

View Details

Steve @dbstev

3 months ago

@dhh @intel @Dell Shit... about to buy a new laptop then!!

0 0 0 20 0

View Details

Steve @dbstev

3 months ago

@cathrynlavery Is it worth it though? I quit openclaw because baby sitting it defeats the purpose completely, and if we have to patch it like crazy to make it "work", maybe better to build something else...