what is a multimodal LLM thinking as it watches a video?
Gemma 4 12B reads raw image patches, as if they were tokens. It was never trained to predict anything at these 'tokens' - but this video shows what it would predict if you did sample from its next token prediction head
use the word `cutover` for clean departures from previous api surface when working with LLMs, without it they have a tendency to unnecessarily preserve backwards compatibility
can't find the original post that taught me this :(
@rfleury@theitiev@fzen_t Contig is guaranteed for the OS virtual address space but that memory can still be phys out of order, which is fine because the contiguousness in virt address space rather than physical space was the goal? Helps the address table lookups, can physical contig also be requested?
@thsottiaux When @ mentioning files in the cli and theres like 20+ matches it only shows the first 7 and doesnt let me downarrow to keep revealing them