ollama / ollama

Get up and running with Llama 3, Mistral, Gemma, and other large language models.

Home Page:https://ollama.com

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Ram not releasing

Stampsm opened this issue · comments

What is the issue?

I seem to have an issue with ram not releasing after model unloads. I waited for the new version with the PS command and verified that the model was unloaded after the default 5 minute timeout and a large duffer time after that just to be safe, but my ram usage was still showing ~60GB on this fresh install of Ubuntu the exact same as it was before the model unloaded. When I try to again load a large model after a while when it's loading my system will crash from what seems like an out of memory issues as this proxmox VM image only has ~120GB allocated to it. Nothing else is running on this image as it was a fresh install just for Ollama and openwebui interface. Is there any suggestions or info needed?

OS

Linux

GPU

Other

CPU

Intel

Ollama version

0.1.38

I just had this again. When I load a larger model and it shows unloaded a few hours later then try to load a different large model my whole system soon crashes. Is there a log someplace I can review?

Can you share a little more data? How are you measuring RAM usage before/during/after loading models? Could you run something like ps -eo pid,ppid,rss,args | grep ollama demonstrating what you're seeing?

If you're still having trouble, please share more info about your measurements and I'll re-open the issue. You can find our troubleshooting docs here https://github.com/ollama/ollama/blob/main/docs/troubleshooting.md