Hermes agent hits compression errors after model switch
worked correctly when Groq’s free Llama 3.3 70B Versatile model was first tested. After it was set as the , with through OpenRouter as the fallback, errors began at the 4 a.m. .
New sessions immediately showed compression errors and said the was exceeded by about 14 to 16 tokens. Llama has a smaller , but the error appeared even at the start of a new session. Switching models during a session produced a warning that the was shrinking.
The size had not changed since the successful test. Turning compression off caused a separate “no compression” error, and changing the compression model to did not fix the problem.
Key points
- Groq’s Llama 3.3 70B Versatile worked in an initial test.
- Compression errors began after the model became the default at a .
- Fresh sessions still reported the as slightly too long.
- Turning compression off or moving compression to did not solve it.
- Agent builders should test and compression settings before relying on a cheaper model.