Salesforce / instructcodet5p-16b

InstructCodeT5+ 16B

Model description

CodeT5+ is a new family of open code large language models with an encoder-decoder architecture that can flexibly operate in different modes (i.e. encoder-only , decoder-only , and encoder-decoder ) to support a wide range of code understanding and generation tasks. It is introduced in the paper:

CodeT5+: Open Code Large Language Models for Code Understanding and Generation by Yue Wang *, Hung Le *, Akhilesh Deepak Gotmare , Nghi D.Q. Bui , Junnan Li , Steven C.H. Hoi (* indicates equal contribution).

Compared to the original CodeT5 family (base: 220M , large: 770M ), CodeT5+ is pretrained with a diverse set of pretraining tasks including span denoising , causal language modeling , contrastive learning , and text-code matching to learn rich representations from both unimodal code data and bimodal code-text data. Additionally, it employs a simple yet effective compute-efficient pretraining method to initialize the model components with frozen off-the-shelf LLMs such as CodeGen to efficiently scale up the model (i.e. 2B , 6B , 16B ), and adopts a "shallow encoder and deep decoder" architecture. Furthermore, it is instruction-tuned to align with natural language instructions (see our InstructCodeT5+ 16B) following Code Alpaca .

How to use

This model can be easily loaded using the AutoModelForSeq2SeqLM functionality and employs the same tokenizer as CodeGen .

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

checkpoint = "Salesforce/instructcodet5p-16b"
device = "cuda" # for GPU usage or "cpu" for CPU usage

tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint,
                                              torch_dtype=torch.float16,
                                              low_cpu_mem_usage=True,
                                              trust_remote_code=True).to(device)

encoding = tokenizer("def print_hello_world():", return_tensors="pt").to(device)
encoding['decoder_input_ids'] = encoding['input_ids'].clone()
outputs = model.generate(**encoding, max_length=15)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Pretraining data

This checkpoint is trained on the stricter permissive subset of the deduplicated version of the github-code dataset . The data is preprocessed by reserving only permissively licensed code ("mit" “apache-2”, “bsd-3-clause”, “bsd-2-clause”, “cc0-1.0”, “unlicense”, “isc”). Supported languages (9 in total) are as follows: c , c++ , c-sharp , go , java , javascript , php , python , ruby.

Training procedure

This checkpoint is initialized from off-the-shelf LLMs, i.e. its encoder is initialized from CodeGen-350M-mono and its decoder is initialized from CodeGen-16B-mono . It is trained on the unimodal code data at the first-stage pretraining, which includes a diverse set of pretraining tasks including span denoising and two variants of causal language modeling . After that, it is further trained on the Python subset with the causal language modeling objective for another epoch to better adapt for Python code generation. Finally, we apply instruction tuning to align it with natural language instructions following Code Alpaca .
Please refer to the paper for more details.

Evaluation results

CodeT5+ models have been comprehensively evaluated on a wide range of code understanding and generation tasks in various settings: zero-shot , finetuning , and instruction-tuning . Specifically, CodeT5+ yields substantial performance gains on many downstream tasks compared to their SoTA baselines, e.g., 8 text-to-code retrieval tasks (+3.2 avg. MRR), 2 line-level code completion tasks (+2.1 avg. Exact Match), and 2 retrieval-augmented code generation tasks (+5.8 avg. BLEU-4). In 2 math programming tasks on MathQA-Python and GSM8K-Python, CodeT5+ models of below billion-parameter sizes significantly outperform many LLMs of up to 137B parameters. Particularly, in the zero-shot text-to-code generation task on HumanEval benchmark, InstructCodeT5+ 16B sets new SoTA results of 35.0% pass@1 and 54.5% pass@10 against other open code LLMs, even surpassing the closed-source OpenAI code-cushman-001 mode Please refer to the paper for more details.

BibTeX entry and citation info

@article{wang2023codet5plus,
  title={CodeT5+: Open Code Large Language Models for Code Understanding and Generation},
  author={Wang, Yue and Le, Hung and Gotmare, Akhilesh Deepak and Bui, Nghi D.Q. and Li, Junnan and Hoi, Steven C. H.},
  journal={arXiv preprint},
  year={2023}
}