Help with later setup steps

[kchatzinikolaou]

So, I managed to download odysseus fine, after some struggles I also finally managed to get a single AI model, because the downloads kept crashing so I had to do everything manually through powershell. Now that I have Qwen3.5-2B-MTP running, it keeps stopping every few seconds, and then no matter what I type back it will keep on handling the first prompt. Is there a reason for this? Can I fix it myself or is it a bug?

Overall it feels like the model isn't processing prompts correctly...
I'm using Native Windown build, not through a docker

[BarsatZulkarnine]

Could you share some more context? A screenshot or any logs from the Odysseus console would help narrow this down. Also, how much RAM does your machine have and is the model running on CPU or GPU?

Once we have that, probably can give you a better answer but for now its worth noting:

Context window filling up - Qwen3.5-2B has a small default context window in Ollama. When it fills, the model stalls or resets. Try running ollama run qwen3.5:2b in Powershell and set num_ctx to something lower like 2048. You can also set this in Odysseus under Settings > Advanced.

RAM pressure - A 2B model is small but on Windows, Ollama may be paging in/out of memory. Check Task Manager while it runs to see if RAM or swap is spiking.
The 'stuck on the first prompt' part is a known Windows issue being worked on in PR #1697

(#1697). It is caused by a reconnect loop that does not exist on Windows, making it look like the first request is still running. A workaround for now is to refresh the page between messages.