Hi, currently we do not test which is the best way for code embedding. We would suggest that you can directly employ the last decoder hidden state or the max/avg pool of all decoder states as the code embedding.
Hi,
Is there any update on evaluating good embeddings of code?
Can you suggest the best possible embedding that could be used to cluster a large number of code-snippets to identify common defects among them?
Hi, currently we do not test which is the best way for code embedding. We would suggest that you can directly employ the last decoder hidden state or the max/avg pool of all decoder states as the code embedding.