metadata

library_name: transformers
license: mit
datasets:
  - jhu-clsp/jfleg
language:
  - en
base_model:
  - google-t5/t5-base
pipeline_tag: text2text-generation

📚 Model Card for Grammar Correction Model

This is a grammar correction model based on the Google T5 architecture, fine-tuned on the JHU-CLSP/JFLEG dataset for text correction tasks. ✍️

Model Details

This model is designed to correct grammatical errors in English sentences. It was fine-tuned using the JFLEG dataset, which provides examples of grammatically correct sentences.

Follow the Developer: Abdul Samad Siddiqui (@samadpls) 👨‍💻

Uses

This model can be directly used to correct grammar and spelling mistakes in sentences. ✅

Example Usage

Here's a basic code snippet to demonstrate how to use the model:

from transformers import T5ForConditionalGeneration, T5Tokenizer

# Load the model and tokenizer
model_name = "samadpls/t5-base-grammar-checker"
tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)

# Example input
example_1 = "grammar: This sentences, has bads grammar and spelling!"

# Tokenize and generate corrected output
inputs = tokenizer.encode(example_1, return_tensors="pt")
outputs = model.generate(inputs)
corrected_sentence = tokenizer.decode(outputs[0], skip_special_tokens=True)

print("Corrected Sentence:", corrected_sentence)

Training Details

The model was trained on the JHU CLSP JFLEG dataset, which includes various examples of sentences with grammatical errors and their corrections. 📖

Training Procedure

Training Hardware: Personal laptop with NVIDIA GeForce MX230 GDDR5 and 16GB RAM 💻
Training Time: Approximately 1 hour ⏳
Hyperparameters: No specific hyperparameters were set for training.

Training Logs

Step	Training Loss	Validation Loss
1	0.9282	0.6091
2	0.6182	0.5561
3	0.6279	0.5345
4	0.6345	0.5147
5	0.5636	0.5076
6	0.6009	0.4928
7	0.5469	0.4950
8	0.5797	0.4834
9	0.5619	0.4818
10	0.6342	0.4788
11	0.5481	0.4786

Final Training Metrics

Training Runtime: 1508.2528 seconds ⏱️
Training Samples per Second: 1.799
Training Steps per Second: 0.225
Final Training Loss: 0.5925
Final Epoch: 1.0

Model Card Contact

For inquiries, please contact Abdul Samad Siddiqui via GitHub. 📬