I’m truly thankful for your assistance so far. Every piece of advice you’ve provided has been incredibly beneficial!
Have a productive day!
1 Like
For the record, it worked. Nonetheless, I encountered a significant error due to non-compliance with the required training data format, specifically the presence of extra spaces. This oversight led to several hours of confusion, as I struggled to understand why the models were not performing as anticipated. I’m sharing this experience for anyone who might just get started and run into a similar issue, hoping it might provide a solution, albeit in rare cases.
1 Like
I thought after fine tuning I can save the tokens of the system prompt? Use the system prompt like in the dataset increase the input tokens of the fine tuned model a lot.
1 Like
Hi - Unfortunately, if you’ve used a system prompt in your fine-tuning data set, you can’t leave it out when you consume the fine-tuned model otherwise you may not get proper results.
Often, however, you can be more focused in the way you design your system prompt. As the model “learns” by demonstration from the training examples, there are normally opportunities to shorten the system prompt and hence save money.
2 Likes