HackerLangs
TopNewTrendsCommentsPastAskShowJobs

Readerium

no profile record

comments

Readerium
·10 hari yang lalu·discuss
3.5 Pro not yet launched, you mean 3.1 pro?
Readerium
·bulan lalu·discuss
Loll
Readerium
·2 bulan yang lalu·discuss
For MSFT: Just download DeepSeek locally and use it.

Or train your own power efficient stack.
Readerium
·2 bulan yang lalu·discuss
AI workloads are all about memory size and bandwidth not compute
Readerium
·2 bulan yang lalu·discuss
LLMs are memory bandwidth bound not compute bound.
Readerium
·3 bulan yang lalu·discuss
Add an iPad mini esque screen on the trackpad of a MacBook.

Trackpad of a 16 inch MacBook Pro is humongous anyways.

Add a touchscreen display to the trackpad, and give it iPad OS
Readerium
·3 bulan yang lalu·discuss
that is true. gguf does not support any Architecture.

for the most recent example, as of April 16, 2026 (today)

Turboquant isnt still added to GGUF
Readerium
·3 bulan yang lalu·discuss
perhaps increasing repitition_penalty might be helpful
Readerium
·3 bulan yang lalu·discuss
In coding they are worse.

Chinese models (GLM, MiniMax) are better.
Readerium
·4 bulan yang lalu·discuss
Can someone explain if the 3D Vcache are stacked on top of each other or side by side.

If they are stacked then why not 9800X3D2?
Readerium
·4 bulan yang lalu·discuss
Qwen 3.5 4B is the goat then
Readerium
·4 bulan yang lalu·discuss
Yes he is room-mates with.
Readerium
·9 bulan yang lalu·discuss
99.999 percent lol
Readerium
·10 bulan yang lalu·discuss
So was intel in the following sentence.
Readerium
·10 bulan yang lalu·discuss
[dead]