Trainer.evaluate() >> TypeError: vars() argument must have __dict__ attribute

1

When training my model, I get the following error during the evaluation step:

TypeError: vars() argument must have __dict__ attribute

I can recreate this error also by simply running trainer.evaluate(). The model appears functional as I can save then reload it and execute inference successfully. So I’m unsure what would cause the trained.evaluate() to throw this error.

Any recommendations?

2 
target_names=["false", "true"] # [0, 1]
employee_count_df['label_as_int'] = (employee_count_df['label'] == True).astype(int)
labels = employee_count_df.label_as_int.values.tolist()
labels = np.array(labels)
full_dataset = employee_count_df.text.values.tolist()

class TenKDataset(torch.utils.data.Dataset):
    def __init__(self, encodings, labels):
        self.encodings = encodings
        self.labels = labels

    def __getitem__(self, idx):
        item = {k: torch.tensor(v[idx]) for k, v in self.encodings.items()}
        item["labels"] = torch.tensor([self.labels[idx]])
        return item

    def __len__(self):
        return len(self.labels)
    
def compute_metrics(pred):
    labels = pred.label_ids
    preds = pred.predictions.argmax(-1)
    # calculate accuracy using sklearn's function
    acc = accuracy_score(labels, preds)
    return {
      'accuracy': acc,
    }

(train_dataset, validation_dataset,train_labels,validation_labels)=train_test_split(full_dataset, labels, test_size=0.3)

train_encodings = tokenizer(train_dataset, truncation=True, padding=True, max_length=max_length)
train_dataset = TenKDataset(train_encodings, train_labels)

model_directory = "./output"
model=BertForSequenceClassification.from_pretrained(model_name, num_labels=len(target_names))
training_args = TrainingArguments(
    num_train_epochs=3,
    per_device_train_batch_size=8,  
    per_device_eval_batch_size=20,  
    weight_decay=0.01,              
    load_best_model_at_end=True,
    logging_steps=300,
    save_steps=300,
    evaluation_strategy="steps",
    output_dir=model_directory
)

trainer = Trainer(  model=model,
                    args=training_args,
                    train_dataset=train_dataset,
                    eval_dataset=validation_dataset,
                    compute_metrics=compute_metrics,)
trainer.train()
3

I assume you are using some higher-level API and the Trainer class comes from HuggingFace?
If so, I guess this issue might be related.