code-vs-nl / README.md
usvsnsp's picture
Librarian Bot: Add base_model information to model (#2)
962c30d
|
raw
history blame
1.86 kB
metadata
language:
  - en
license: apache-2.0
tags:
  - generated_from_trainer
datasets:
  - bookcorpus
  - codeparrot/github-code
metrics:
  - accuracy
  - f1
base_model: distilbert-base-uncased
model-index:
  - name: code-vs-nl
    results: []

code-vs-nl

This model is a fine-tuned version of distilbert-base-uncased on bookcorpus for text and codeparrot/github-code for code datasets. It achieves the following results on the evaluation set:

  • Loss: 0.5180
  • Accuracy: 0.9951
  • F1 Score: 0.9950

Model description

As it's a finetuned model, it's architecture is same as distilbert-base-uncased for Sequence Classification

Intended uses & limitations

Can be used to classify documents into text and code

Training and evaluation data

It is a mix of above two datasets, equally random sampled

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 1e-07
  • train_batch_size: 256
  • eval_batch_size: 1024
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • training_steps: 1000

Training results

Training Loss Epoch Step Validation Loss Accuracy F1 Score
0.5732 0.07 500 0.5658 0.9934 0.9934
0.5254 0.14 1000 0.5180 0.9951 0.9950

Framework versions

  • Transformers 4.25.1
  • Pytorch 1.13.1+cu116
  • Datasets 2.8.0
  • Tokenizers 0.13.2