josecannete's picture
Update README.md
8740aef
metadata
language:
  - es
tags:
  - albert
  - spanish
  - OpenCENIA
datasets:
  - large_spanish_corpus

ALBERT Large Spanish

This is an ALBERT model trained on a big spanish corpora. The model was trained on a single TPU v3-8 with the following hyperparameters and steps/time:

  • LR: 0.000625
  • Batch Size: 512
  • Warmup ratio: 0.003125
  • Warmup steps: 12500
  • Goal steps: 4000000
  • Total steps: 1450000
  • Total training time (aprox): 42 days.

Training loss

https://drive.google.com/uc?export=view&id=10EiI0Yge3U3CnGrqoMs1yJY020pPz_Io