|
https: |
|
from transformers import AutoModelForSeq2SeqLM |
|
from peft import get_peft_config, get_peft_model, LoraConfig, TaskType |
|
model_name_or_path = "bigscience/mt0-large" |
|
tokenizer_name_or_path = "bigscience/mt0-large" |
|
|
|
peft_config = LoraConfig( |
|
task_type=TaskType.SEQ_2_SEQ_LM, inference_mode=False, r=8, lora_alpha=32, lora_dropout=0.1 |
|
) |
|
|
|
model = AutoModelForSeq2SeqLM.from_pretrained(model_name_or_path) |
|
model = get_peft_model(model, peft_config) |
|
model.print_trainable_parameters() |
|
"trainable params: 2359296 || all params: 1231940608 || trainable%: 0.19151053100118282" |
|
@Misc{peft, |
|
title = {PEFT: State-of-the-art Parameter-Efficient |
|
author |
|
howpublished ///huggingface/ |
|
year |
|
} |