Model stops responding after some chat messages

[weigher12]

After upgrading openwebui to the newest version I have this issue that the model stops responding after some messages.
with a thinking model like gemma4:12b it just says "thinking for 10s" and no response after that.
with gemma3:12b the streaming response just stops in the middle.

If I increase the context lengh in ollama settings from 8k to 16k, it works just fine.
so is it linked to the context lenght ?

[Classic298]

yes context length

[youdie006]

Yes, this is very likely related to context length, especially since increasing the Ollama context from 8k to 16k makes the same chat work.

A few things are happening at once:

• The model sees the full conversation history, system prompt, model parameters, and sometimes hidden/reasoning content within the available context window.
• Reasoning/thinking models can consume a lot of tokens before producing the final answer, so they hit the limit earlier than a non-reasoning model.
• When the chat gets close to or beyond num_ctx, Ollama/Open WebUI may appear to stop mid-stream or get stuck after “thinking”, because the model no longer has enough usable room to continue cleanly.

What I would try:

1. Keep num_ctx at 16k or higher for that model if your hardware can handle it.
2. Start a fresh chat with the same prompt. If the fresh chat works but the long chat fails, it is almost certainly context pressure.
3. Shorten or summarize older messages in long conversations.
4. Check the Ollama/Open WebUI logs when the stream stops. If you see context/token or memory pressure messages, that confirms it.
5. If GPU/RAM usage gets too high after increasing context, lower num_ctx again and use shorter chats/summaries instead.

So the short answer is: yes, it is probably linked to context length, but also watch memory usage because larger context windows cost much more memory.