I’m wondering if more frequent fine tuning with less data, say 10 times with 100 examples, yields better results than single fine-tuning with more data, say once with 1000 examples. Anyone has any info about that?
Also I just fine tuned an already fine tuned model. And it went smooth, no errors. Can I be sure that my initial training is not lost? I.e. my second training just added to the initial training?
Anyone experienced with fine tuning please help.