Am using a NanoGPT sub and tried various models from there. If I use 'Quick Start' then the game runs fine (eg using the DeepSeek R1 model). But if I try 'New Game', picking the students manually, it just hangs for ages (10+ mins) at 'Generating Class Schedule' and eventually comes back with a 'generation failed' type message. This is whatever model I've tried - smaller ones, large ones, coding-focused, RP-focused, etc. I know probably you'll say 'Use Gemini' but shouldn't one of these other models also be capable of reasonable JSON or whatever it is you're looking for in the responses? Why does it take so long to fail, also? Any debug I could look at to narrow down what's going wrong?
Viewing post in v0.0.1 - Known Bugs, feedback, suggestions
You can download an app.log from the main menu by clicking Feedback (or checking data/app.log). Usual culprits for generation failed are the model not returning unique class IDs for each class. Deepseek R1 shouldn't have a problem unless you're running it locally though, that's weird. On Gemini Flash 3.8, the prompts finish in under 30s