Hi,
It looks like the repository contains the adapter weights.
To load the entire model (base model with adapter weights), you can use the AutoPeftModelForSeq2Seq class:
from peft import AutoPeftModelForSeq2SeqLM
base_model_id = "facebook/nllb-200-distilled-600M"
adapter_model_id = "your-adapter-repo"
peft_model = AutoPeftModelForSeq2SeqLM.from_pretrained(base_model_id, adapter_model_id)
# now you can call the merge_and_unload method which returns a regular NLLB model which you can use for inference
model = peft_model.merge_and_unload()