Loading Lora models after trainning

Beginners
1

Hello guys, i am facing difficulties saving and LoRa models.

here are my codes,

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, DataCollatorForSeq2Seq, BitsAndBytesConfig

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type='nf4',
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True
)

model_name = "google/mt5-base"
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name, quantization_config=bnb_config)
data_collator = DataCollatorForSeq2Seq(tokenizer, model=model, padding=True, return_tensors="pt")

from peft import LoraConfig, get_peft_model, LoraModel, prepare_model_for_kbit_training

config = LoraConfig(
    task_type="SEQ_2_SEQ_LM",
    r=16,
    lora_alpha=16,
    target_modules=["q", "v"],
    lora_dropout=0.1,
    bias="none",
)
lora_model = LoraModel(model, config, "default")
peft_model = prepare_model_for_kbit_training(model)
model.add_adapter(config)
model.save_pretrained("./models/mt5_quant")

from transformers import Seq2SeqTrainer, Seq2SeqTrainingArguments

training_args = Seq2SeqTrainingArguments(
    output_dir="./model/mt5_quant_ipa/",
    group_by_length=True,
    length_column_name="length",
    per_device_train_batch_size=4,
    per_device_eval_batch_size=4,
    gradient_accumulation_steps=16,
    gradient_checkpointing=True,
    evaluation_strategy="steps",
    metric_for_best_model="wer",
    greater_is_better=False,
    load_best_model_at_end=True,
    num_train_epochs=5,
    save_steps=500,
    eval_steps=500,
    logging_steps=1000,
    learning_rate=3e-4,
    weight_decay=1e-2,
    warmup_steps=1000,
    save_total_limit=5,
    predict_with_generate=True,
    generation_max_length=512,
    push_to_hub=False,
    bf16=True,
    tf32=True,
    optim="adafactor",
)

trainer = Seq2SeqTrainer(
    model=model,
    tokenizer=tokenizer,
    args=training_args,
    data_collator=data_collator,
    train_dataset=ds_train,
    eval_dataset=ds_val,
    compute_metrics=compute_metrics,
)

The problem arises when the trainer tries to load a model from check point, it throws an error that ‘pytorch_model.bin’ not found while trying to load from checkpoint-XXXX or somehting like that (sorry should have copied the error)

how to solve this? i tried reading the docs but i a barely understood.

Thank you!

2

config = PeftConfig.from_pretrained(save_dir)
model = AutoModelForSequenceClassification.from_pretrained(save_dir)
model = PeftModel.from_pretrained(eval_model, save_dir)

try to load it like this