3 ms·
I do this for many application. 2 to 4 RTXA5000 do the job (Lora finetune). As for dataset, depending on your task, you need image / text pairs.
by pilooch 2y ago
I do this for many application. 2 to 4 RTXA5000 do the job (Lora finetune). As for dataset, depending on your task, you need image / text pairs.
- magicalhippo 2y ago> As for dataset, depending on your task, you need image / text pairs. I guess the main question is, do you just prepare training data as if you were training from scratch, or is there some particularities to finetuning that should be considered?