Commit
•
cba44cb
1
Parent(s):
02cb712
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,6 +1,8 @@
|
|
| 1 |
---
|
| 2 |
license: cc-by-nc-4.0
|
| 3 |
base_model: Helsinki-NLP/opus-mt-tc-big-en-ar
|
|
|
|
|
|
|
| 4 |
model-index:
|
| 5 |
- name: Terjman-Large-v2
|
| 6 |
results: []
|
|
@@ -8,17 +10,16 @@ datasets:
|
|
| 8 |
- atlasia/darija_english
|
| 9 |
language:
|
| 10 |
- ar
|
|
|
|
| 11 |
---
|
| 12 |
|
| 13 |
-
#
|
| 14 |
-
|
| 15 |
-
This model is trained to translate English text (en) into Moroccan Darija text (Ary) written in Arabic letters.
|
| 16 |
-
|
| 17 |
-
## Model Overview
|
| 18 |
|
| 19 |
Our model is built upon the powerful Transformer architecture, leveraging state-of-the-art natural language processing techniques.
|
| 20 |
-
It has been finetuned on a the
|
|
|
|
| 21 |
|
|
|
|
| 22 |
|
| 23 |
## Training hyperparameters
|
| 24 |
|
|
@@ -34,13 +35,6 @@ The following hyperparameters were used during training:
|
|
| 34 |
- lr_scheduler_warmup_ratio: 0.03
|
| 35 |
- num_epochs: 30
|
| 36 |
|
| 37 |
-
## Framework versions
|
| 38 |
-
|
| 39 |
-
- Transformers 4.39.2
|
| 40 |
-
- Pytorch 2.2.2+cpu
|
| 41 |
-
- Datasets 2.18.0
|
| 42 |
-
- Tokenizers 0.15.2
|
| 43 |
-
|
| 44 |
## Usage
|
| 45 |
|
| 46 |
Using our model for translation is simple and straightforward.
|
|
@@ -66,7 +60,7 @@ output_tokens = model.generate(**input_tokens)
|
|
| 66 |
# Decode the output tokens
|
| 67 |
output_text = tokenizer.decode(output_tokens[0], skip_special_tokens=True)
|
| 68 |
|
| 69 |
-
print("
|
| 70 |
```
|
| 71 |
|
| 72 |
## Example
|
|
@@ -86,4 +80,11 @@ We're currently collecting more data with the aim of continous improvements.
|
|
| 86 |
## Feedback
|
| 87 |
|
| 88 |
We're continuously striving to improve our model's performance and usability and we will be improving it incrementaly.
|
| 89 |
-
If you have any feedback, suggestions, or encounter any issues, please don't hesitate to reach out to us.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: cc-by-nc-4.0
|
| 3 |
base_model: Helsinki-NLP/opus-mt-tc-big-en-ar
|
| 4 |
+
metrics:
|
| 5 |
+
- bleu
|
| 6 |
model-index:
|
| 7 |
- name: Terjman-Large-v2
|
| 8 |
results: []
|
|
|
|
| 10 |
- atlasia/darija_english
|
| 11 |
language:
|
| 12 |
- ar
|
| 13 |
+
- en
|
| 14 |
---
|
| 15 |
|
| 16 |
+
# Terjman-Large-v2 (240M params)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
Our model is built upon the powerful Transformer architecture, leveraging state-of-the-art natural language processing techniques.
|
| 19 |
+
It has been finetuned on a the [darija_english](atlasia/darija_english) dataset enhanced with curated corpora ensuring high-quality and accurate translations.
|
| 20 |
+
This model is an impovement of the previous version [Terjman-Large](atlasia/Terjman-Large).
|
| 21 |
|
| 22 |
+
The finetuning was conducted using a **A100-40GB** and took **17 hours**.
|
| 23 |
|
| 24 |
## Training hyperparameters
|
| 25 |
|
|
|
|
| 35 |
- lr_scheduler_warmup_ratio: 0.03
|
| 36 |
- num_epochs: 30
|
| 37 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
## Usage
|
| 39 |
|
| 40 |
Using our model for translation is simple and straightforward.
|
|
|
|
| 60 |
# Decode the output tokens
|
| 61 |
output_text = tokenizer.decode(output_tokens[0], skip_special_tokens=True)
|
| 62 |
|
| 63 |
+
print("Translation:", output_text)
|
| 64 |
```
|
| 65 |
|
| 66 |
## Example
|
|
|
|
| 80 |
## Feedback
|
| 81 |
|
| 82 |
We're continuously striving to improve our model's performance and usability and we will be improving it incrementaly.
|
| 83 |
+
If you have any feedback, suggestions, or encounter any issues, please don't hesitate to reach out to us.
|
| 84 |
+
|
| 85 |
+
## Framework versions
|
| 86 |
+
|
| 87 |
+
- Transformers 4.39.2
|
| 88 |
+
- Pytorch 2.2.2+cpu
|
| 89 |
+
- Datasets 2.18.0
|
| 90 |
+
- Tokenizers 0.15.2
|