Please clone the CodeCSE repository to get GraphCodeBERTForCL and other dependencies to use this pretrained model. https://github.com/emu-se/CodeCSE

Detailed instructions are listed in the repository's README.md. Overall, you will need:

GraphCodeBERT (CodeCSE uses GraphCodeBERT's input format for code)
GraphCodeBERTForCL defined in codecse/codecse

Inference example

NL input example: example_nl.json

json

{
    "original_string": "", 
    "docstring_tokens": ["Save", "model", "to", "a", "pickle", "located", "at", "path"], 
    "url": "https://github.com/openai/baselines/blob/3301089b48c42b87b396e246ea3f56fa4bfc9678/baselines/deepq/deepq.py#L55-L72"
}

Code snippet to get the embedding of an NL document (link to complete code):

nl_json = load_example("example_nl.json")
batch = prepare_inputs(nl_json, tokenizer, args)
nl_inputs = batch[3]
with torch.no_grad():        
    nl_vec = model(input_ids=nl_inputs, sent_emb="nl")[1]

sjiang1/codecse

Model Card for CodeCSE

Inference example

sjiang1/
codecse