---
language:
- es
tags:
- albert
- spanish
- OpenCENIA
datasets:
- large_spanish_corpus
---

# ALBERT XXLarge Spanish

This is an [ALBERT](https://github.com/google-research/albert) model trained on a [big spanish corpora](https://github.com/josecannete/spanish-corpora).
The model was trained on a single TPU v3-8 with the following hyperparameters and steps/time:
- LR: 0.0003125
- Batch Size: 128
- Warmup ratio: 0.00078125
- Warmup steps: 3125
- Goal steps: 4000000
- Total steps: 1650000
- Total training time (aprox): 70.7 days.

## Training loss

![https://drive.google.com/uc?export=view&id=1a9MHsk-QwBuCMtyDyRvZ5mv9Mzl2dWCn](https://drive.google.com/uc?export=view&id=1a9MHsk-QwBuCMtyDyRvZ5mv9Mzl2dWCn)