README.md

metadata

license: mit
tags:
  - code

What does this model do?

This model converts the natural language input to Neo4j (Cypher) query. It is a fine-tuned CodeT5+ 220M. This model is a part of nl2query repository which is present at https://github.com/Chirayu-Tripathi/nl2query

You can use this model via the github repository or via following code. More information can be found on the repository.

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
model = AutoModelForSeq2SeqLM.from_pretrained("Chirayu/nl2cql")
tokenizer = AutoTokenizer.from_pretrained("Chirayu/nl2cql")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)

textual_query = '''cypher: find the outcomes of people who are female and below the age of 32 | case : gender, reportdate, ageunit, reporteroccupation, primaryid, age, eventDate | outcome : code, outcome | relationships : resulted_in'''

def generate_query(
        textual_query: str,
        num_beams: int = 10,
        max_length: int = 128,
        repetition_penalty: int = 2.5,
        length_penalty: int = 1,
        early_stopping: bool = True,
        top_p: int = 0.95,
        top_k: int = 50,
        num_return_sequences: int = 1,
    ) -> str:
        input_ids = tokenizer.encode(
            textual_query, return_tensors="pt", add_special_tokens=True
        )
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        input_ids = input_ids.to(device)
        generated_ids = model.generate(
            input_ids=input_ids,
            num_beams=num_beams,
            max_length=max_length,
            repetition_penalty=repetition_penalty,
            length_penalty=length_penalty,
            early_stopping=early_stopping,
            top_p=top_p,
            top_k=top_k,
            num_return_sequences=num_return_sequences,
        )
        query = [
            tokenizer.decode(
                generated_id,
                skip_special_tokens=True,
                clean_up_tokenization_spaces=True,
            )
            for generated_id in generated_ids
        ][0]
        return query