Server reruns same task multiple times

I used
```
deploy = HuggingFaceModel(
name=model_name,
role=role,
code_location="abc",
model_data=path_to_s3,
transformers_version="4.37",
pytorch_version="2.1",
py_version='py310',
model_server_workers=1,
)
emb = deploy.deploy(
endpoint_name=model_name,
initial_instance_count=1,
instance_type="ml.c5.4xlarge",
container_startup_health_check_timeout=300,
)
```
The custom script was
```
def model_fn(model_dir):
processor = DataProcess() # A class that contains logic for processing each file.
return processor

def predict_fn(data, model):
text = model.process_file(data)
return {"output": text}
```
The input `data` is a base64 string of a file content.
It's strange that when the file is pretty small, under 1MB, the server runs `model_fn` and `predict_fn` once, and the process took around 30 seconds. But when I inputted large file of around 1.5MB, it runs `model_fn` and `predict_fn` multiple times, each time lasting around 2mins. I know this because the same request gives multiple contents of
```
[INFO ] W-model-1-stdout com.amazonaws.ml.mms.wlm.WorkerLifeCycle - Preprocess time - 5.128383636474609 ms
[INFO ] W-model-1-stdout com.amazonaws.ml.mms.wlm.WorkerLifeCycle - Predict time - 162199.17178153992 ms
[INFO ] W-model-1-stdout com.amazonaws.ml.mms.wlm.WorkerLifeCycle - Postprocess time - 0.00762939453125 ms
```

It's probably unorthodox to use the server for the data processing job. But what configs did I miss?

Related: https://github.com/aws/amazon-sagemaker-examples/issues/1073

Start with the HuggingFaceModel deploy call and the model_fn and predict_fn entry points shown in the report. Reproduce the contrast between a sub-1MB and approximately 1.5MB base64 request while inspecting worker lifecycle logs, then determine whether request handling or a missing configuration causes repeated execution; done means the cause and applicable configuration are documented and the request runs once.

Written by the indexing model from the issue text.