RDF format for aws Neptune

I am not an RDFLIB expert, but as N-Triples format is highly parallelizable and assuming your CSV format is the same (e.g., every row is independent), I would suggest looking to parallelize your process. If you split the file into say 20 files of 1,000 and run them in parallel, whether as a Glue job, AWS Batch job, or maybe containers...or even kicking off 20 shell processes on an EC2 instance, beyond doing the transformation in parallel, it will also likely be faster because it will have less memory pressure.