MT5-es-to-quy

This model was trained using synthetic data across several iterations of State-of-the-art Spanish to Quechua (Y) datasets.

Model Details

Model Description

The model presents a Test Bleu: 2.0391 and Test chrF: 22.6785 which was tested into another dataset (contest). Our validation results (Synthetic) were Bleu: 11.0126 and Validation chrF: 31.8344.

• Developed by:
  • Julio Santisteban Pablo
  • Ricardo Lazo Vasquez
• Shared by [optional]: Julio Santisteban Pablo
• Model type: Translate
• Language(s) (NLP): Spanish to Quechua (Y)
• License: Apache 2.0
• FinetuFine-tunedmodel: MT5

Model Sources

• Repository: Julio/mt5-es-to-quy Uses

  The overall goal is to improve translation from Spanish to Quechua(Y) and their related applications.

  Direct Use

  Improve translations from spanish to Quechua (Y).

  Downstream Use

  Create High-impact platforms who helps Quechua speakers.

  Out-of-Scope Use

  • Generate malicious content.
  • A 100% trustable platform.

  Bias, Risks, and Limitations

  The overall bias in this model is that is a work in progress, which is not recommended to use this model for a critical scenario.

  Recommendations

  It is important to evaluate the model in your own data before use it blindly. If the end user thinks is necesary an additional training, it must be done in the same conditions over the expected productive data.

  How to Get Started with the Model

  You can use it normally importing the model using MT5 as a base.

  Training Details

  Training Data

  The model was trained using syntetic data from different sources. The overall dataset will be published in the future and this notes will be updated accordingly.

  Training Procedure

  The model was trained over the syntetic dataset and tested in a contest dataset. (In the future, we will give more details).

  Preprocessing

  The model was preprocessed several times for each attempt of final model delivery. (In the future, we will give more details).

  Training Hyperparameters

  (In the future, we will give more details).

  Speeds, Sizes, Times

  (In the future, we will give more details).

  Evaluation

  Testing Data, Factors & Metrics

  Testing Data

  (In the future, we will give more details).

  Factors

  The main end user is the speakers of Quechua (Y) dialect.

  Metrics

  We evaluate our model using BLEU and cHrF evaluation metrics.

  Results

  The model presents a Test Bleu: 2.0391 and Test chrF: 22.6785 which was tested into another dataset (contest). Our validation results (Synthetic) were Bleu: 11.0126 and Validation chrF: 31.8344.

  Summary

  This work was made by several researchers from universities from Peru.

  (In the future, we will give more details).

  Model Examination

  (In the future, we will give more details).

  Environmental Impact

  Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

  Technical Specifications

  Model Architecture and Objective

  (In the future, we will give more details).

  Compute Infrastructure

  (In the future, we will give more details).

  Hardware

  (In the future, we will give more details).

  Software

  The model was trained using PyTorch.

  Citation

  (In the future, we will give more details).

  More Information

  Please feel free to mail to the main researcher Julio Santisteban to his email: jsantisteban@ucsp.edu.pe.

  Model Card Authors

  Ricardo Lazo Vasquez

  Model Card Contact

  Please email us to:

  • jsantisteban@ucsp.edu.pe
  • CC: ricardo.lazo@ucsp.edu.pe