ONNXRuntime CPU - Memory spiking continuously (Memory leak)

Hi,

Describe the issue
I am using ONNX Runtime python api for inferencing, during which the memory is spiking continuosly.

(Model information - Converted pytorch based transformers model to ONNX and quantized it)

Urgency
Critical

System information

• ONNX Runtime version: 1.8.1
• Python version: 3.6
• Running on a CPU backend

Inference code being used

For creating onnxruntime session:

from onnxruntime import InferenceSession, GraphOptimizationLevel, SessionOptions

options = SessionOptions()
options.intra_op_num_threads = 1
options.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL
session = InferenceSession(model_path, options, providers=["CPUExecutionProvider"])
session.disable_fallback()

for sentence in sentence_list:                   # sentence_list = list of sentences
	ort_inputs = tokenizer(sentence, truncation=True, return_tensors="pt")
	ort_inputs = {
		'input_ids': ort_inputs.data['input_ids'].numpy(),
		'attention_mask': ort_inputs.data['attention_mask'].numpy()
	}
	ort_outputs_quant = session.run(None, ort_inputs)[0]        # onnx_model = instance of loaded onnxmodel 
	pred_logits.extend(ort_outputs_quant)
	del ort_outputs_quant
	del ort_inputs
y_pred = np.argmax(pred_logits, axis=1)

• Also tried with latest onnxruntime == 1.9.0 with python 3.7. The issue still persists.

Below is the code used to export DistilRobertaBase Pytorch model to ONNX and quantize it.

# Get the example data to run the model and export it to ONNX
enable_overwrite = True
inputs = {
    'input_ids':      input_ids[0].to(device).reshape(1, max_seq_length), 
    'attention_mask': attention_masks[0].to(device).reshape(1, max_seq_length)
}

model.eval()
model.to(device)

# Converting PyTorch model to ONNX format
if enable_overwrite or not os.path.exists(export_model_path):
    with torch.no_grad():
        symbolic_names = {0: 'batch_size', 1: 'max_seq_len'}
        torch.onnx.export(model,                                            # model being run
                          args=tuple(inputs.values()),                      # model input (or a tuple for multiple inputs)
                          f=export_model_path,                              # where to save the model (can be a file or file-like object)
                          opset_version=11,                                 
                          do_constant_folding=True,                         # whether to execute constant folding for optimization
                          input_names=['input_ids',                         # the model's input names
                                       'attention_mask'],
                          output_names=['output_0'],                    # the model's output names
                          dynamic_axes={'input_ids': symbolic_names,        # variable length axes
                                        'attention_mask' : symbolic_names,
                                        'output_0' : {0: 'batch_size'}})

In above code, example value for variable inputs is:

{
    'input_ids': tensor([[    0, 44879,  4576,    35,  1437,  1437,   508,  2940,  2890,  1225,
           5046,  2481,     2,     1,     1,     1,     1,     1,     1,     1]]),
    'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0]])
}

Erroneous behavior

Memory keeps spiking during run.

Expected behavior
Expecting stable memory consumption throughout the run (no memory leaks or accumulation).