Update (ver_240813)
Browse files
README.md
CHANGED
@@ -7,6 +7,7 @@ datasets:
|
|
7 |
- tatoeba
|
8 |
metrics:
|
9 |
- bleu
|
|
|
10 |
model-index:
|
11 |
- name: opus_model
|
12 |
results:
|
@@ -22,7 +23,11 @@ model-index:
|
|
22 |
metrics:
|
23 |
- name: Bleu
|
24 |
type: bleu
|
25 |
-
value:
|
|
|
|
|
|
|
|
|
26 |
---
|
27 |
|
28 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
@@ -32,11 +37,11 @@ should probably proofread and complete it, then remove this comment. -->
|
|
32 |
|
33 |
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-ja-pl](https://huggingface.co/Helsinki-NLP/opus-mt-ja-pl) on the tatoeba dataset.
|
34 |
It achieves the following results on the evaluation set:
|
35 |
-
- Loss: 1.
|
36 |
-
- Bleu:
|
37 |
-
- Gen Len: 9.
|
38 |
-
- Meteor: 0.
|
39 |
-
- Chrf:
|
40 |
|
41 |
## Model description
|
42 |
|
@@ -56,42 +61,37 @@ Training with kaggle notebook (GPU) on GPU P100.
|
|
56 |
|
57 |
The following hyperparameters were used during training:
|
58 |
- learning_rate: 3e-05
|
59 |
-
- train_batch_size:
|
60 |
- eval_batch_size: 8
|
61 |
- seed: 42
|
62 |
-
- optimizer:
|
63 |
- lr_scheduler_type: linear
|
64 |
-
- num_epochs:
|
65 |
- mixed_precision_training: Native AMP
|
66 |
|
67 |
### Examples
|
68 |
|
69 |
-
| Japanese | Original translation | DeepL | Opus-mt-ja-pl-
|
70 |
|---------------------------|----------------------|------ |-------------------|
|
71 |
| 今ちょっとやることがあってね | Mam teraz coś do zrobienia. | Mam teraz kilka rzeczy do zrobienia. | Mam teraz kilka spraw do załatwienia. |
|
72 |
| なぜッあの少女を助けてやらなかったのだ! | Czemu jej nie pomogłeś! | Dlaczego nie pomogłeś tej dziewczynie? | Dlaczego jej nie pomogłeś?! |
|
73 |
| ここで何をしている? | Czego tu szukacie? | Co ty tu robisz? | Co tu robisz? |
|
74 |
| あんたの協力が要る | Potrzebujemy cię. | Potrzebuję twojej pomocy. | Potrzebuję twojej pomocy. |
|
75 |
-
| こたえはなに? | A jaka jest właściwie odpowiedź? | Jaka jest odpowiedź? | Co
|
76 |
-
| 一人で寝んのが怖くなったんか? | Boisz się spać sama? | Boisz się spać samotnie? | Boisz się spać
|
77 |
|
78 |
### Training results
|
79 |
|
80 |
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len | Meteor | Chrf |
|
81 |
|:-------------:|:-----:|:------:|:---------------:|:-------:|:-------:|:------:|:-------:|
|
82 |
-
| 2.5658 | 1.0 |
|
83 |
-
|
|
84 |
-
|
|
85 |
-
| 2.1139 | 4.0 | 226724 | 1.2833 | 28.9288 | 9.4581 | 0.5244 | 49.4667 |
|
86 |
-
| 1.9825 | 5.0 | 283405 | 1.2170 | 31.3751 | 9.3229 | 0.5358 | 51.0005 |
|
87 |
-
| 1.8982 | 6.0 | 340086 | 1.1660 | 32.9805 | 9.4976 | 0.5563 | 52.5487 |
|
88 |
-
| 1.8198 | 7.0 | 396767 | 1.1305 | 34.0223 | 9.4436 | 0.5665 | 53.2912 |
|
89 |
-
| 1.7592 | 8.0 | 453448 | 1.1164 | 34.4952 | 9.442 | 0.5692 | 53.728 |
|
90 |
|
91 |
|
92 |
### Framework versions
|
93 |
|
94 |
-
- Transformers 4.
|
95 |
-
- Pytorch 2.
|
96 |
- Datasets 2.20.0
|
97 |
-
- Tokenizers 0.19.1
|
|
|
7 |
- tatoeba
|
8 |
metrics:
|
9 |
- bleu
|
10 |
+
- chrf
|
11 |
model-index:
|
12 |
- name: opus_model
|
13 |
results:
|
|
|
23 |
metrics:
|
24 |
- name: Bleu
|
25 |
type: bleu
|
26 |
+
value: 37.84
|
27 |
+
language:
|
28 |
+
- pl
|
29 |
+
- ja
|
30 |
+
library_name: transformers
|
31 |
---
|
32 |
|
33 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
|
|
37 |
|
38 |
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-ja-pl](https://huggingface.co/Helsinki-NLP/opus-mt-ja-pl) on the tatoeba dataset.
|
39 |
It achieves the following results on the evaluation set:
|
40 |
+
- Loss: 1.6326
|
41 |
+
- Bleu: 37.8457
|
42 |
+
- Gen Len: 9.2006
|
43 |
+
- Meteor: 0.589
|
44 |
+
- Chrf: 0.589
|
45 |
|
46 |
## Model description
|
47 |
|
|
|
61 |
|
62 |
The following hyperparameters were used during training:
|
63 |
- learning_rate: 3e-05
|
64 |
+
- train_batch_size: 12
|
65 |
- eval_batch_size: 8
|
66 |
- seed: 42
|
67 |
+
- optimizer: adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08
|
68 |
- lr_scheduler_type: linear
|
69 |
+
- num_epochs: 12
|
70 |
- mixed_precision_training: Native AMP
|
71 |
|
72 |
### Examples
|
73 |
|
74 |
+
| Japanese | Original translation | DeepL | Opus-mt-ja-pl-240813 |
|
75 |
|---------------------------|----------------------|------ |-------------------|
|
76 |
| 今ちょっとやることがあってね | Mam teraz coś do zrobienia. | Mam teraz kilka rzeczy do zrobienia. | Mam teraz kilka spraw do załatwienia. |
|
77 |
| なぜッあの少女を助けてやらなかったのだ! | Czemu jej nie pomogłeś! | Dlaczego nie pomogłeś tej dziewczynie? | Dlaczego jej nie pomogłeś?! |
|
78 |
| ここで何をしている? | Czego tu szukacie? | Co ty tu robisz? | Co tu robisz? |
|
79 |
| あんたの協力が要る | Potrzebujemy cię. | Potrzebuję twojej pomocy. | Potrzebuję twojej pomocy. |
|
80 |
+
| こたえはなに? | A jaka jest właściwie odpowiedź? | Jaka jest odpowiedź? | Co masz na myśli? |
|
81 |
+
| 一人で寝んのが怖くなったんか? | Boisz się spać sama? | Boisz się spać samotnie? | Boisz się spać samemu? |
|
82 |
|
83 |
### Training results
|
84 |
|
85 |
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len | Meteor | Chrf |
|
86 |
|:-------------:|:-----:|:------:|:---------------:|:-------:|:-------:|:------:|:-------:|
|
87 |
+
| 2.5658 | 1.0 | 57000 | 1.6196 | 22.1938 | 9.341 | 0.4576 | 44.3828 |
|
88 |
+
| 1.8982 | 6.0 | 343170 | 1.9423 | 31.1109 | 9.2355 | 0.5389 | 51.7878 |
|
89 |
+
| 1.6991 | 11.0 | 629145 | 1.1164 | 37.8457 | 9.2006 | 0.589 | 56.6614 |
|
|
|
|
|
|
|
|
|
|
|
90 |
|
91 |
|
92 |
### Framework versions
|
93 |
|
94 |
+
- Transformers 4.44.0
|
95 |
+
- Pytorch 2.4.0
|
96 |
- Datasets 2.20.0
|
97 |
+
- Tokenizers 0.19.1
|