# 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/codegen2-3_7B

pip install hf-hub-ctranslate2>=2.0.8 

Converted on 2023-05-21 using

ct2-transformers-converter --model Salesforce/codegen2-3_7B --output_dir /home/michael/tmp-ct2fast-codegen2-3_7B --force --copy_files merges.txt tokenizer.json README.md tokenizer_config.json vocab.json special_tokens_map.json added_tokens.json configuration_codegen.py .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-codegen2-3_7B"
# 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/codegen2-3_7B")
)
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

CodeGen2 (CodeGen2-3.7B)

Model description

CodeGen2 is a family of autoregressive language models for program synthesis, introduced in the paper:

CodeGen2: Lessons for Training LLMs on Programming and Natural Languages by Erik Nijkamp*, Hiroaki Hayashi*, Caiming Xiong, Silvio Savarese, Yingbo Zhou.

Unlike the original CodeGen model family (i.e., CodeGen1), CodeGen2 is capable of infilling, and supports more programming languages.

Four model sizes are released: 1B, 3.7B, 7B, 16B.

How to use

This model can be easily loaded using the AutoModelForCausalLM functionality.

Causal sampling

For regular causal sampling, simply generate completions given the context:

from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("Salesforce/codegen2-3_7B")
model = AutoModelForCausalLM.from_pretrained("Salesforce/codegen2-3_7B", trust_remote_code=True, revision="main")

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))

Infill sampling

For infill sampling, we introduce three new special token types:

• <mask_N>: N-th span to be masked. In practice, use <mask_1> to where you want to sample infill.
• <sep>: Seperator token between the suffix and the infilled sample. See below.
• <eom>: "End-Of-Mask" token that model will output at the end of infilling. You may use this token to truncate the output.

For example, if we want to generate infill for the following cursor position of a function:

def hello_world():
    |
    return name

we construct an input to the model by

1. Inserting <mask_1> token in place of cursor position
2. Append <sep> token to indicate the boundary
3. Insert another <mask_1> to indicate which mask we want to infill.

The final snippet looks as follows:

from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("Salesforce/codegen2-3_7B")
model = AutoModelForCausalLM.from_pretrained("Salesforce/codegen2-3_7B", trust_remote_code=True, revision="main")


def format(prefix, suffix):
  return prefix + "<mask_1>" + suffix + "<|endoftext|>" + "<sep>" + "<mask_1>"


prefix = "def hello_world():\n    "
suffix = "    return name"
text = format(prefix, suffix)
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=False)[len(text):])

You might want to truncate the model output with <eom>.

Training data

This checkpoint is trained on the stricter permissive subset of the deduplicated version of the Stack dataset (v1.1). Supported languages (and frameworks) are as follows: c, c++, c-sharp, dart, go, java, javascript, kotlin, lua, php, python, ruby, rust, scala, shell, sql, swift, typescript, vue.

Training procedure

CodeGen2 was trained using cross-entropy loss to maximize the likelihood of sequential inputs. The input sequences are formatted in two ways: (1) causal language modeling and (2) file-level span corruption. Please refer to the paper for more details.

Evaluation results

We evaluate our models on HumanEval and HumanEval-Infill. Please refer to the paper for more details.

Intended use and limitations

As an autoregressive language model, CodeGen2 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.

BibTeX entry and citation info

@article{Nijkamp2023codegen2,
  title={CodeGen2: Lessons for Training LLMs on Programming and Natural Languages},
  author={Nijkamp, Erik and Hayashi, Hiroaki and Xiong, Caiming and Savarese, Silvio and Zhou, Yingbo},
  journal={arXiv preprint},
  year={2023}
}