ValueError: Unable to create tensor, you should probably activate padding with 'padding=True' - nlp

Question

I am trying to evaluate facebook/hubert-base-ls9601 Huggingface pre-trained model after fine-tuning on a private dataset.
I am using facebook/hubert-base-ls9601 pre-trained model, and Wav2vec2 feature extractor, and pooling mode set to mean.
Here's the evaluation code:
test_dataset = load_dataset("csv", data_files={"test": "/content/drive/MyDrive/freelancing/test.csv"}, delimiter="\t")["test"]
def speech_file_to_array_fn(batch):
speech_array, sampling_rate = torchaudio.load(batch["path"])
resampler = torchaudio.transforms.Resample(sampling_rate, target_sampling_rate)
speech = resampler(speech_array).squeeze().numpy()
batch["speech"] = speech_array
return batch
def predict(batch):
features = feature_extractor(batch["speech"], sampling_rate=feature_extractor.sampling_rate, return_tensors="pt", padding=True)
input_values = features.input_values.to(device)
with torch.no_grad():
logits = model(input_values).logits
pred_ids = torch.argmax(logits, dim=-1).detach().cpu().numpy()
batch["predicted"] = pred_ids
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
result = test_dataset.map(predict, batched=True, batch_size=2)
On the last line of code, I encounter the following error block:
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
/usr/local/lib/python3.7/dist-packages/transformers/feature_extraction_utils.py in convert_to_tensors(self, tensor_type)
168 if not is_tensor(value):
--> 169 tensor = as_tensor(value)
170
ValueError: could not broadcast input array from shape (2,220683) into shape (2,)
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
12 frames
<ipython-input-73-7bd88adad349> in <module>()
----> 1 result = test_dataset.map(predict, batched=True, batch_size=2)
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in map(self, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, num_proc, suffix_template, new_fingerprint, desc)
1970 new_fingerprint=new_fingerprint,
1971 disable_tqdm=disable_tqdm,
-> 1972 desc=desc,
1973 )
1974 else:
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
517 self: "Dataset" = kwargs.pop("self")
518 # apply actual function
--> 519 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
520 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
521 for dataset in datasets:
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in wrapper(*args, **kwargs)
484 }
485 # apply actual function
--> 486 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
487 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
488 # re-apply format to the output
/usr/local/lib/python3.7/dist-packages/datasets/fingerprint.py in wrapper(*args, **kwargs)
456 # Call actual function
457
--> 458 out = func(self, *args, **kwargs)
459
460 # Update fingerprint of in-place transforms + update in-place history of transforms
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in _map_single(self, function, with_indices, with_rank, input_columns, batched, batch_size, drop_last_batch, remove_columns, keep_in_memory, load_from_cache_file, cache_file_name, writer_batch_size, features, disable_nullable, fn_kwargs, new_fingerprint, rank, offset, disable_tqdm, desc, cache_only)
2340 indices,
2341 check_same_num_examples=len(input_dataset.list_indexes()) > 0,
-> 2342 offset=offset,
2343 )
2344 except NumExamplesMismatchError:
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in apply_function_on_filtered_inputs(inputs, indices, check_same_num_examples, offset)
2217 if with_rank:
2218 additional_args += (rank,)
-> 2219 processed_inputs = function(*fn_args, *additional_args, **fn_kwargs)
2220 if update_data is None:
2221 # Check if the function returns updated examples
/usr/local/lib/python3.7/dist-packages/datasets/arrow_dataset.py in decorated(item, *args, **kwargs)
1912 )
1913 # Use the LazyDict internally, while mapping the function
-> 1914 result = f(decorated_item, *args, **kwargs)
1915 # Return a standard dict
1916 return result.data if isinstance(result, LazyDict) else result
<ipython-input-71-6f845da29c00> in predict(batch)
11
12 def predict(batch):
---> 13 features = feature_extractor(batch["speech"], sampling_rate=feature_extractor.sampling_rate, return_tensors="pt", padding=True)
14
15 input_values = features.input_values.to(device)
/usr/local/lib/python3.7/dist-packages/transformers/models/wav2vec2/feature_extraction_wav2vec2.py in __call__(self, raw_speech, padding, max_length, truncation, pad_to_multiple_of, return_attention_mask, return_tensors, sampling_rate, **kwargs)
200 truncation=truncation,
201 pad_to_multiple_of=pad_to_multiple_of,
--> 202 return_attention_mask=return_attention_mask,
203 )
204
/usr/local/lib/python3.7/dist-packages/transformers/feature_extraction_sequence_utils.py in pad(self, processed_features, padding, max_length, truncation, pad_to_multiple_of, return_attention_mask, return_tensors)
230 batch_outputs[key].append(value)
231
--> 232 return BatchFeature(batch_outputs, tensor_type=return_tensors)
233
234 def _pad(
/usr/local/lib/python3.7/dist-packages/transformers/feature_extraction_utils.py in __init__(self, data, tensor_type)
78 def __init__(self, data: Optional[Dict[str, Any]] = None, tensor_type: Union[None, str, TensorType] = None):
79 super().__init__(data)
---> 80 self.convert_to_tensors(tensor_type=tensor_type)
81
82 def __getitem__(self, item: str) -> Union[Any]:
/usr/local/lib/python3.7/dist-packages/transformers/feature_extraction_utils.py in convert_to_tensors(self, tensor_type)
174 raise ValueError("Unable to create tensor returning overflowing values of different lengths. ")
175 raise ValueError(
--> 176 "Unable to create tensor, you should probably activate padding "
177 "with 'padding=True' to have batched tensors with the same length."
178 )
ValueError: Unable to create tensor, you should probably activate padding with 'padding=True' to have batched tensors with the same length.
I am working on Google Colab. Those are the environment variables:
%env LC_ALL=C.UTF-8
%env LANG=C.UTF-8
%env TRANSFORMERS_CACHE=/content/cache
%env HF_DATASETS_CACHE=/content/cache
%env CUDA_LAUNCH_BLOCKING=1
The padding is already activated in the predict function.
Can you please help me fix it?