CodeRosetta/CodeRosetta_cpp2cuda_ft

1---2language: en3license: apache-2.04---5 6# CodeRosetta7## Pushing the Boundaries of Unsupervised Code Translation for Parallel Programming ([📃Paper](https://arxiv.org/abs/2410.20527), [🔗Website](https://coderosetta.com/)).8 9 10CodeRosetta is an EncoderDecoder translation model. It supports the translation of C++, CUDA, and Fortran. \11This version of the model is fine-tuned on synthetic dataset for **C++ to CUDA translation.**12 13### How to use14 15```python16from transformers import AutoTokenizer, EncoderDecoderModel17 18# Load the CodeRosetta model and tokenizer19model = EncoderDecoderModel.from_pretrained('CodeRosetta/CodeRosetta_cpp2cuda_ft')20tokenizer = AutoTokenizer.from_pretrained('CodeRosetta/CodeRosetta_cpp2cuda_ft')21 22# Encode the input C++ Code23input_cpp_code = "void add_100 ( int numElements , int * data ) { for ( int idx = 0 ; idx < numElements ; idx ++ ) { data [ idx ] += 100 ; } }"24input_ids = tokenizer.encode(input_cpp_code, return_tensors="pt")25 26# Set the start token to <CUDA>27start_token = "<CUDA>"28decoder_start_token_id = tokenizer.convert_tokens_to_ids(start_token)29 30# Generate the CUDA code31output = model.generate(32    input_ids=input_ids, 33    decoder_start_token_id=decoder_start_token_id,34    max_length=25635)36 37# Decode and print the generated output38generated_code = tokenizer.decode(output[0], skip_special_tokens=True)39print(generated_code)40```41 42### BibTeX 43 44```bibtex45@inproceedings{coderosetta:neurips:2024,46  title = {CodeRosetta: Pushing the Boundaries of Unsupervised Code Translation for Parallel Programming},47  author = {TehraniJamsaz, Ali and Bhattacharjee, Arijit and Chen, Le and Ahmed, Nesreen K and Yazdanbakhsh, Amir and Jannesari, Ali},48  booktitle = {NeurIPS},49  year = {2024},50}51