# Fast-Inference with Ctranslate2

Speedup inference while reducing memory by 2x-4x using int8 inference in C++ on CPU or GPU.

quantized version of Salesforce/codegen-6B-mono

pip install hf-hub-ctranslate2>=2.0.8

Converted on 2023-05-21 using

ct2-transformers-converter --model Salesforce/codegen-6B-mono --output_dir /home/michael/tmp-ct2fast-codegen-6B-mono --force --copy_files merges.txt tokenizer.json README.md tokenizer_config.json vocab.json special_tokens_map.json added_tokens.json .gitattributes --quantization float16

Checkpoint compatible to ctranslate2>=3.13.0 and hf-hub-ctranslate2>=2.0.6

compute_type=int8_float16 for device="cuda"
compute_type=int8 for device="cpu"

from hf_hub_ctranslate2 import TranslatorCT2fromHfHub, GeneratorCT2fromHfHub
from transformers import AutoTokenizer

model_name = "michaelfeil/ct2fast-codegen-6B-mono"
# use either TranslatorCT2fromHfHub or GeneratorCT2fromHfHub here, depending on model.
model = GeneratorCT2fromHfHub(
        # load in int8 on CUDA
        model_name_or_path=model_name, 
        device="cuda",
        compute_type="int8_float16",
        # tokenizer=AutoTokenizer.from_pretrained("Salesforce/codegen-6B-mono")
)
outputs = model.generate(
    text=["def print_hello_world():", "def hello_name(name:"],
    max_length=64
)
print(outputs)

Licence and other remarks:

This is just a quantized version. Licence conditions are intended to be idential to original huggingface repo.

Original description

tags:

ctranslate2
int8
float16

CodeGen (CodeGen-Mono 6B)

Model description

CodeGen is a family of autoregressive language models for program synthesis from the paper: A Conversational Paradigm for Program Synthesis by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong. The models are originally released in this repository, under 3 pre-training data variants (NL, Multi, Mono) and 4 model size variants (350M, 2B, 6B, 16B).

The checkpoint included in this repository is denoted as CodeGen-Mono 6B in the paper, where "Mono" means the model is initialized with CodeGen-Multi 6B and further pre-trained on a Python programming language dataset, and "6B" refers to the number of trainable parameters.

Training data

This checkpoint (CodeGen-Mono 6B) was firstly initialized with CodeGen-Multi 6B, and then pre-trained on BigPython dataset. The data consists of 71.7B tokens of Python programming language. See Section 2.1 of the paper for more details.

Training procedure

CodeGen was trained using cross-entropy loss to maximize the likelihood of sequential inputs. The family of models are trained using multiple TPU-v4-512 by Google, leveraging data and model parallelism. See Section 2.3 of the paper for more details.

Evaluation results

We evaluate our models on two code generation benchmark: HumanEval and MTPB. Please refer to the paper for more details.

Intended Use and Limitations

As an autoregressive language model, CodeGen is capable of extracting features from given natural language and programming language texts, and calculating the likelihood of them. 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 as well.

How to use

This model can be easily loaded using the AutoModelForCausalLM functionality:

from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("Salesforce/codegen-6B-mono")
model = AutoModelForCausalLM.from_pretrained("Salesforce/codegen-6B-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}
}