File size: 710 Bytes
c6a1f78 cb442f7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 |
---
language:
- es
tags:
- albert
- spanish
- OpenCENIA
datasets:
- large_spanish_corpus
---
# ALBERT XLarge 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: 6250
- Goal steps: 8000000
- Total steps: 2775000
- Total training time (aprox): 64.2 days.
## Training loss
 |