Very slow transcription time on second message

This is kind of a specific one. I’m in Windows, and I had Llama 3.3 70B as my AI model in Groq, but I kept hitting a limit, so I switched to Llama 3.1 8B Instant which works great, processes fast… the first time I use it.

If I transcribe something, and then start another one relatively quickly (anywhere under 10s roughly), then the second transcription processing takes way longer, 30s to a minute.