Finetuning
#10
by
kaidanti
- opened
I was trying to finetune the model, but kept running into issues with training examples being skipped:This instance will be ignored in loss calculation. Note, if this happens often, consider increasing the
max_seq_length.
I increased the value up to 128_000 but still keep running into this issue.
What are you using for FT? Here is a ref implementation from the Meta team:
https://github.com/meta-llama/llama-recipes/tree/main/recipes/quickstart/finetuning
found the issue; I am using the DataCollatorForCompletionOnlyLM
and had the wrong templates
kaidanti
changed discussion status to
closed
Thanks! Looking forward to hearing about your experiments :)