mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-10 15:57:04 +00:00
use a similar strategy as llama.cpp for deciding where tensors should be allocated. this will be improved later to be aware of usable memory before assigning the tensor