license: mit
datasets:
- ammarnasr/the-stack-java-clean
library_name: adapter-transformers
tags:
- code
pipeline_tag: text-generation
language:
- code
CodeGen (CodeGen-Mono 350M LoRa Java)
Model description
CodeGen LoRa Java is a family of autoregressive language models fine-tuned using LoRa on Different Programming Langauges.
Training data
This model was fine-tuned on the cleaned Java subset from TheStack Avilable here. The data consists of 1 Million Java code files.
Training procedure
This model was fine-tuned using LoRa on 1 T4 GPU. The model was trained for 10,000 steps with batch size of 4. The model was trained using causal language modeling loss.
Evaluation results
We evaluate our models on the MultiPle-E bencchmark. The model achieves 8.9 Pass@10 Rate.
Intended Use and Limitations
However, the model is intended for and best at program synthesis, that is, generating executable code given English prompts, where the prompts should be in the form of a comment string. The model can complete partially-generated code in Java and Python.
How to use
This model can be easily loaded using the AutoModelForCausalLM functionality:
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("ammmarnasr/codegen-350M-mono-java")
model = AutoModelForCausalLM.from_pretrained("Salesforce/codegen-350M-mono")
text = "def hello_world():"
input_ids = tokenizer(text, return_tensors="pt").input_ids
generated_ids = model.generate(input_ids, max_length=128)
print(tokenizer.decode(generated_ids[0], skip_special_tokens=True))
BibTeX entry and citation info
@article{Nijkamp2022ACP,
title={A Conversational Paradigm for Program Synthesis},
author={Nijkamp, Erik and Pang, Bo and Hayashi, Hiroaki and Tu, Lifu and Wang, Huan and Zhou, Yingbo and Savarese, Silvio and Xiong, Caiming},
journal={arXiv preprint},
year={2022}
}