Commit
•
f6a2543
1
Parent(s):
08f8865
Update README.md
Browse files
README.md
CHANGED
@@ -9,7 +9,7 @@ tags:
|
|
9 |
- PyTorch
|
10 |
- Safetensors
|
11 |
widget:
|
12 |
-
- text: 'translate
|
13 |
---
|
14 |
|
15 |
# mt5-large-spanish-nahuatl
|
@@ -19,6 +19,14 @@ Nahuatl is the most widely spoken indigenous language in Mexico, yet training a
|
|
19 |
## Model description
|
20 |
This model is an MT5 Transformer ([mt5-large](https://huggingface.co/google/mt5-large)) fine-tuned on Spanish and Nahuatl sentences collected from diverse places online. The dataset is normalized using 'inali' normalization from [py-elotl](https://github.com/ElotlMX/py-elotl).
|
21 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
## Usage
|
24 |
```python
|
@@ -71,12 +79,12 @@ Also, additional 30,000 samples were collected from the web to enhance the data.
|
|
71 |
The employed method uses a single training stage using the mt5. This model was leveraged given that it can handle different vocabularies and prefixes.
|
72 |
|
73 |
### Training
|
74 |
-
The model is trained till convergence, adding the prefixes "translate spanish to nahuatl: + word" and "translate nahuatl to spanish: + word".
|
|
|
75 |
|
76 |
### Training setup
|
77 |
The model uses the same dataset for 77,500 steps using batch size = 4 and a learning rate of 1e-4.
|
78 |
|
79 |
-
|
80 |
## Evaluation results
|
81 |
The models are evaluated on 2 different datasets:
|
82 |
1. First on the test sentences similar to the evaluation ones.
|
|
|
9 |
- PyTorch
|
10 |
- Safetensors
|
11 |
widget:
|
12 |
+
- text: 'translate nahuatl to spanish: Nimitstlazohkamate'
|
13 |
---
|
14 |
|
15 |
# mt5-large-spanish-nahuatl
|
|
|
19 |
## Model description
|
20 |
This model is an MT5 Transformer ([mt5-large](https://huggingface.co/google/mt5-large)) fine-tuned on Spanish and Nahuatl sentences collected from diverse places online. The dataset is normalized using 'inali' normalization from [py-elotl](https://github.com/ElotlMX/py-elotl).
|
21 |
|
22 |
+
## Inference API use
|
23 |
+
You can change and translate to nahuatl or spanish just do:
|
24 |
+
|
25 |
+
translate spanish to nahuatl: Quiero agua
|
26 |
+
|
27 |
+
or
|
28 |
+
|
29 |
+
translate nahuatl to spanish: Nimitstlazohkamate
|
30 |
|
31 |
## Usage
|
32 |
```python
|
|
|
79 |
The employed method uses a single training stage using the mt5. This model was leveraged given that it can handle different vocabularies and prefixes.
|
80 |
|
81 |
### Training
|
82 |
+
The model is trained bidirectionally till convergence, adding the prefixes "translate spanish to nahuatl: + word" and "translate nahuatl to spanish: + word".
|
83 |
+
This is meant as an improvement of [previous models](https://huggingface.co/hackathon-pln-es/t5-small-spanish-nahuatl).
|
84 |
|
85 |
### Training setup
|
86 |
The model uses the same dataset for 77,500 steps using batch size = 4 and a learning rate of 1e-4.
|
87 |
|
|
|
88 |
## Evaluation results
|
89 |
The models are evaluated on 2 different datasets:
|
90 |
1. First on the test sentences similar to the evaluation ones.
|