While I'm having much more success getting OpenHan...
# questionss
straight-magician-16269
12/12/2025, 10:49 PMWhile I'm having much more success getting OpenHands started on programming apps of different kinds, I'm still running into issues where it continually loops for hours and hours and eventually gives up and spits out: `Timeout: litellm.Timeout: APITimeoutError - Request timed out. Error_str: Request timed out.`Does this sound like exhausting the context window? and if so, is there a way to continue developing past the first few hundred LOC?
c
calm-artist-7206
12/13/2025, 2:06 AMA Timeout can be just that the LLM API didn't respond, it can happen randomly.
If it was context window, OpenHands has a mechanism to condense after a number of events. Was the condenser enabled in your run?
calm-artist-7206
12/13/2025, 2:07 AMI don't see in the log any "condensation event", so it didn't happen
calm-artist-7206
12/13/2025, 2:08 AMA random timeout shouldn't matter, you could try to just continue, as I think you did
calm-artist-7206
12/13/2025, 2:09 AMI see your LLM was Qwen-3-coder 30b
calm-artist-7206
12/13/2025, 2:10 AMWe have evaluated a Qwen3-coder 480b, and it seemed very good on software tasks:
https://docs.google.com/spreadsheets/d/1wOUdFCMyY6Nt0AIqF705KN4JKOWgeI4wUGUP60krXXs/edit?gid=0#gid=0
calm-artist-7206
12/13/2025, 2:10 AMits bigger brother 😊
s
straight-magician-16269
12/13/2025, 3:56 AMyeah I can certainly try 480b,, but im only on 12gb VRAM. I had it happen again on my most recent attempt and when i asked it to try again it ran the condenser immediately
straight-magician-16269
12/13/2025, 4:07 AMI can try to clear a bunch of HD space for 480b and give it a try overnight but I don't have particularly high hopes lol, LM Studio says likely too large to run. It's a shame there is nothing in between. Right now openhands is just spinning in circles trying to load a python module it created and failing every time with ModuleNotFound errors :(
straight-magician-16269
12/13/2025, 4:12 AMAnd now its failing with context window overflow so I'm even more skeptical that a larger model will have better results since my 12gb VRAM will probably provide a very small context-window
c
calm-artist-7206
12/13/2025, 4:22 AMwhen i asked it to try again it ran the condenser immediatelyAh, sounds like you might be right, it timed out when sending and processing half the history, trying to summarize it
s
straight-magician-16269
12/13/2025, 4:23 AMI played around with the settings of the 30b and got it to load with the full 2500k contexxt window so fingers crossed 🙂
straight-magician-16269
12/13/2025, 4:24 AM250k* sorry mechanical kbd
straight-magician-16269
12/13/2025, 4:27 AMI had to take the KV Cache off VRAM but glad I can at least try it
3 Views