Does PEFT support translation models such as M2M100?

[zzzul9]

I'm tring to applied LoRA to M2M100.

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
from peft import get_peft_config, get_peft_model, LoraConfig, TaskType

peft_config = LoraConfig(task_type = TaskType.SEQ_2_SEQ_LM, inference_mode = False, target_modules=["q_proj", "v_proj"], r = 16, lora_alpha = 32, lora_dropout = 0.1)

model_repo = 'facebook/m2m100_418M'
model = AutoModelForSeq2SeqLM.from_pretrained(model_repo, proxies=proxies)
tokenizer = AutoTokenizer.from_pretrained(model_repo, proxies=proxies)

model = get_peft_model(model, peft_config)
model.print_trainable_parameter()

...

peft_model_id="results"
model.save_pretrained(peft_model_id)
tokenizer.save_pretrained(peft_model_id)
torch.save(model.state_dict(), path)

But I got below error message.

AttributeError: 'M2M100ForConditionalGeneration' object has no attribute 'print_trainable_parameter'

If comment out 'print_trainable_parameter()', it runs without error, but I can't be sure if it goes right. Please let me know how to solve it? and if it supports other models not specified in models-support-matrix.

Thank you!

[pacman100]

the function is print_trainable_parameters and not print_trainable_parameter

[pacman100]

[zzzul9]

Thank you for confirming that it is applicable to other models. Also about print_trainable_parameter, sorry for bothering you with a minor mistake.