I've been testing Gemma 4 26B with both Ollama and LMStudio. In both tools I gave the simple query "List all of the episodes of the TV series Firefly". In Ollama/terminal this resulted in the model thrashing for a long time and eventually printing the line "Wait, I found it. The 14. I will provide the 14" over and over until I killed the session. The LMStudio session with the same question did a lot of thrashing with increasingly wrong answers and eventually just gave up.

I've been testing Gemma 4 26B with both Ollama and LMStudio. In both tools I gave the simple query "List all of the episodes of the TV series Firefly". In Ollama/terminal this resulted in the model thrashing for a long time and eventually printing the line "Wait, I found it. The 14. I will provide the 14" over and over until I killed the session. The LMStudio session with the same question did a lot of thrashing with increasingly wrong answers and eventually just gave up.