I am trying to use the AST model for fine-tuning on my dataset and I don’t have any problems until I get to the training step.
This is the error That is being returned
***** Running training *****
Num examples = 2
Num Epochs = 4
Instantaneous batch size per device = 32
Total train batch size (w. parallel, distributed & accumulation) = 128
Gradient Accumulation steps = 4
Total optimization steps = 4
Number of trainable parameters = 86187264
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
<ipython-input-26-97a81c88cb1e> in <module>
----> 1 train_results = trainer.train()
2
3 # save tokenizer with the model
4 trainer.save_model()
5 trainer.log_metrics('train', train_results.metrics)
6 frames
/usr/local/lib/python3.8/dist-packages/transformers/feature_extraction_utils.py in __getitem__(self, item)
84 return self.data[item]
85 else:
---> 86 raise KeyError("Indexing with integers is not available when using Python based feature extractors")
87
88 def __getattr__(self, item: str):
KeyError: 'Indexing with integers is not available when using Python based feature extractors'
And this is my setup
model_path = "MIT/ast-finetuned-audioset-10-10-0.4593"
feature_extractor = ASTFeatureExtractor()
def preprocess_audio(features, labels):
inputs = feature_extractor(features, return_tensors = 'pt', sampling_rate=16000)
inputs['label'] = labels
return inputs
# right now, the tensor is loaded on the cpu. If a GPU is available, we want to load it onto that for training
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# the data train and test are pandas dataframes with a features and labels column
vision_train = preprocess_audio(train['features'].tolist(), train['labels'].tolist())
vision_test = preprocess_audio(test['features'].tolist(), test['labels'].tolist())
def collate(batch):
return {
'input_values': torch.stack([x['input_values'] for x in batch]),
'labels': torch.tensor([x['label'] for x in batch])
}
def compute_metric(p):
return metric.compute(
predictions = np.argmax(p.predictions, axis = 1),
references = p.label_ids
)
# arguments used by the model
args = TrainingArguments(
output_dir = '/content/ast',
per_device_train_batch_size=32,
evaluation_strategy='steps',
gradient_accumulation_steps=4,
num_train_epochs=4,
save_steps = 100,
eval_steps=100,
logging_steps=10,
learning_rate = 3e-5,
warmup_ratio=0.1,
save_total_limit = 2,
remove_unused_columns = False,
push_to_hub = False,
metric_for_best_model="accuracy",
load_best_model_at_end = True
)
trainer = Trainer(
model = model,
args = args,
data_collator = collate,
compute_metrics=compute_metric,
train_dataset = vision_train,
eval_dataset = vision_test,
tokenizer = feature_extractor
)
Happy to provide more context but I’m not really sure what the error is here 
help highly appreciated @stevhliu stevhliu