Problem "Target size must be the same as input size "

[JauLamine]

Hello everyone, I’m encountering an issue with my code for the NLI classification task. I have three labels, and I’m using BERT. When attempting to run my program, I encounter the following error message: “raise ValueError (f"Target size ({target.size()}) must be the same as input size.” Thank you in advance for any assistance. ({input.size()})“)
ValueError: Target size (torch.Size([1])) must be the same as input size (torch.Size([1, 3]))”

-----------------------------------code-------------------------------------------------------------------

-------------Function----------------------

def compute_metrics(eval_pred):
logits = eval_pred.predictions[0]
labels = eval_pred.label_ids
predictions = np.argmax(logits, axis=-1)
return {“f1”: f1_score(labels, predictions)}

def tokenize_function(example):
give_premise = 'Premise: ’
give_hypothesis = "\nHypothesis: "
give_classification = '\n\nClassification: ’
prompt = [give_premise + premise + give_hypothesis + hypothesis + give_classification for premise, hypothesis in zip(example[“premise”], example[“hypothesis”])]
example[‘input_ids’] = tokenizer(prompt, padding=“max_length”, truncation=True, return_tensors=“pt”).input_ids

return example

#--------------Loading dataset----------------

#utilFunction.balanced_dataset(“…/train.csv”, 200, “balanced_train.csv”)
#utilFunction.balanced_dataset(“…/validation_matched.csv”,20, “balanced_validation_matched.csv”)
#utilFunction.balanced_dataset(“…/validation_mismatched.csv”,20, “balanced_validation_mismatched.csv”)

data_files = {“train”: “balanced_train.csv”, “test”: “balanced_validation_matched.csv”, “validation”: “balanced_validation_mismatched.csv”}

#dataset = load_dataset(“csv”, data_files=data_files)

df = pd.read_csv(‘balanced_train.csv’)
df = df.rename(columns={‘NliClass’: ‘labels’})
df[‘labels’] = df[‘labels’].astype(int)
dataset = datasets.Dataset.from_pandas(df)
print(dataset)
label2id = {0: ‘ENTAILMENT’, 1: ‘NEUTRAL’, 2: ‘CONTRADICTION’}
id2label = {‘ENTAILMENT’: 0, ‘NEUTRAL’: 1, ‘CONTRADICTION’: 2}

#--------------Loading Model And Tokenizer----------------

model_name=“bert-base-uncased”
original_model = BertForSequenceClassification.from_pretrained(model_name, problem_type=“multi_label_classification”, num_labels=3,
id2label=id2label, label2id=label2id, ignore_mismatched_sizes=True)

print(original_model)
print(original_model.config.id2label)
tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenized_datasets = dataset.map(tokenize_function, batched=True)

ls = [‘input_ids’, ‘labels’]

for var in df:
if var not in ls:
tokenized_datasets = tokenized_datasets.remove_columns(var)

#--------------Print trainable parameter and generate original output----------------
print(utilFunction.print_number_of_trainable_model_parameters(original_model))

#--------------------- DEFINE-PEFT-LORA---------------------------

lora_config = LoraConfig(
task_type=TaskType.SEQ_CLS, r=1, lora_alpha=1, lora_dropout=0.1
)
model = get_peft_model(original_model, lora_config)

metric = evaluate.load(“accuracy”)

peft_training_args = TrainingArguments(output_dir=“test_trainer”, evaluation_strategy=“epoch”, num_train_epochs=25, per_device_train_batch_size=1)

peft_trainer = Trainer(
model=model,
args=peft_training_args,
train_dataset=tokenized_datasets,
#eval_dataset=tokenized_datasets[“validation”],
compute_metrics=compute_metrics,
)

#--------------------- TRAIN ---------------------------

peft_trainer.train()