UI can be overwhelmed when streaming a huge response that's coming in too fast

[TiagoTiago]

Describe the bug

I just had a spooky moment; I asked something to the LLama 8B distill of Deepseek R1 that made it go into a very long train of thought. After a while I realized it was going in the wrong direction and I needed to reformulate the question to make clear what I was asking about, but the stop button didn't work. I gave it some time, and then when it still didn't stop after additional clicks of the stop button and a little more waiting, I decided to go to the terminal and Ctrl-C the whole thing. But coming back to the browser the damn thing was still "typing" the response!

Only when I reloaded the page it finally showed that I had indeed terminated the process with the Ctrl-C.

Now after I booted up the web UI again, turns out it already had the whole response in the log, so I was actually not waiting the generation to finish, but for the page to finish the "typing animation" of the huge text it already had even on disk.

Is there an existing issue for this?

• I have searched the existing issues

Reproduction

Not sure the exact steps that will guarantee a repro, but the description of the event given above probably has enough clues.

Screenshot

No response

Logs

.

System Info

.