GPT2Coder / README.md
Gerson Fabian Buenahora Ormaza
Update README.md
0776384 verified
|
raw
history blame
1.72 kB
metadata
license: mit
datasets:
  - agicorp/python_code_instructions_18k_alpaca
language:
  - en
base_model:
  - openai-community/gpt2
pipeline_tag: text-generation
library_name: transformers
tags:
  - code

Model Card

GPT2Coder is a language model that uses openAI's GPT2 model architecture, the model was pre-trained on multiple code data focused on python and languages ​​​​such as Spanish and English. The pretrained model was finely tuned to handle the task of receiving textual input in the form of a code request and generating a code output.

Model Details

  • Developed by: BueormAI
  • Shared by: BueormLLC
  • Model type: Transformer
  • Language(s) (NLP): English (en), Spanish (es)
  • License: MiT
  • Finetuned from model: GPT2 Architecture

Bias, Risks, and Limitations

The model can generate unexpected code and output, in addition to offensive texts and non-functional code.

Recommendations

We recommend using the model with caution and handling its outputs with discretion as they may turn out to be non-functional outputs and harmful and dangerous code.

Training Details

Training Hyperparameters

  • Training regime: fp16 mixed precision
  • Max_lenght: 1024 tokens
  • pretrain epochs: 1 epochs
  • finetuning epochs: 2 epochs

Environmental Impact

  • Hardware Type: GPU P100
  • Hours used: 18 hours
  • Cloud Provider: Kaggle

By Bueorm

Thanks to all the people who download and support our projects and manage a vision towards the future with AI, we hope you will support us to continue advancing and launching more followed models.