opus100

Références :

af-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/af-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 275512
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "af",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

amen

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/am-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 89027
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "am",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

an-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/an-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'train' 6961
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "an",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ar-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ar-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ar",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

comme-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/as-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 138479
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "as",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

az-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/az-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 262089
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "az",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

a été

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/be-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 67312
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "be",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

bg-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/bg-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "bg",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

bn-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/bn-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "bn",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

frère

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/br-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 153447
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "br",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

bs-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/bs-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "bs",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ca-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ca-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ca",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

cs-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/cs-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "cs",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

cy-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/cy-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 289521
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "cy",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

bonjour

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/da-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "da",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

de-en

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/de-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "de",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

dz-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/dz-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'train' 624
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "dz",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

el-en

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/el-en')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "el",
            "en"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-eo

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-eo')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 337106
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "eo"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-es

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-es')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "es"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-et

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-et')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "et"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-eu

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-eu')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "eu"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-fa

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-fa')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "fa"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-fi

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-fi')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "fi"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-fr')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "fr"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-fy

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-fy')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 54342
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "fy"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ga

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ga')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 289524
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ga"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-gd

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-gd')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 1606
'train' 16316
'validation' 1605
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "gd"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-gl

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-gl')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 515344
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "gl"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-gu

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-gu')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 318306
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "gu"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ha

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ha')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 97983
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ha"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-il

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-he')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "he"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-salut

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-hi')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 534319
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "hi"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-hr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-hr')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "hr"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-hu

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-hu')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "hu"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-hy

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-hy')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'train' 7059
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "hy"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-id

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-id')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "id"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ig

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ig')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 1843
'train' 18415
'validation' 1843
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ig"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-est

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-is')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "is"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-it

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-it')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "it"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-ja

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ja')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ja"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ka

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ka')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 377306
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ka"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-kk

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-kk')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 79927
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "kk"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-km

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-km')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 111483
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "km"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ko

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ko')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ko"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-kn

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-kn')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 918
'train' 14537
'validation' 917
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "kn"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ku

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ku')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 144844
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ku"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ky

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ky')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 27215
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ky"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-li

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-li')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 25535
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "li"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-lt

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-lt')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "lt"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-lv

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-lv')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "lv"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-mg

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-mg')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 590771
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "mg"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-mk

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-mk')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "mk"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ml

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ml')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 822746
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ml"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-mn

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-mn')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'train' 4294
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "mn"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-mr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-mr')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 27007
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "mr"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ms

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ms')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ms"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-mt

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-mt')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "mt"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-mon

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-my')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 24594
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "my"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-nb

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-nb')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 142906
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "nb"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ne

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ne')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 406381
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ne"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-nl

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-nl')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "nl"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-nn

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-nn')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 486055
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "nn"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-non

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-no')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "no"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-oc

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-oc')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 35791
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "oc"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ou

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-or')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 1318
'train' 14273
'validation' 1317
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "or"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-pa

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-pa')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 107296
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "pa"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-pl

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-pl')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "pl"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ps

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ps')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 79127
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ps"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-pt

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-pt')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "pt"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-ro

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ro')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ro"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ru

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ru')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ru"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-rw

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-rw')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 173823
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "rw"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-se

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-se')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 35907
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "se"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-sh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-sh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 267211
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "sh"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-si

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-si')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 979109
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "si"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-sk

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-sk')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "sk"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-sl

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-sl')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "sl"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-carré

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-sq')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "sq"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-sr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-sr')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "sr"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-sv

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-sv')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "sv"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ta

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ta')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 227014
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ta"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-te

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-te')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 64352
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "te"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-tg

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-tg')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 193882
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "tg"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ième

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-th')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "th"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-tk

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-tk')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 1852
'train' 13110
'validation' 1852
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "tk"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-tr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-tr')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "tr"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-tt

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-tt')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 100843
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "tt"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ug

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ug')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 72170
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ug"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-uk

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-uk')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "uk"
        ],
        "id": null,
        "_type": "Translation"
    }
}

en-ur

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-ur')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 753913
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "ur"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-uz

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-uz')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 173157
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "uz"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-vi

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-vi')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "vi"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-wa

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-wa')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 104496
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "wa"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-xh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-xh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 439671
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "xh"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-yi

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-yi')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 15010
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "yi"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-yo

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-yo')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'train' 10375
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "yo"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-zh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-zh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 1000000
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "zh"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-zu

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/en-zu')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
'train' 38616
'validation' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "en",
            "zu"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ar-de

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ar-de')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ar",
            "de"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ar-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ar-fr')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ar",
            "fr"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ar-nl

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ar-nl')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ar",
            "nl"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ar-ru

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ar-ru')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ar",
            "ru"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ar-zh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ar-zh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ar",
            "zh"
        ],
        "id": null,
        "_type": "Translation"
    }
}

de-fr

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/de-fr')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "de",
            "fr"
        ],
        "id": null,
        "_type": "Translation"
    }
}

de-nl

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/de-nl')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "de",
            "nl"
        ],
        "id": null,
        "_type": "Translation"
    }
}

de-ru

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/de-ru')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "de",
            "ru"
        ],
        "id": null,
        "_type": "Translation"
    }
}

de-zh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/de-zh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "de",
            "zh"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-nl

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/fr-nl')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "fr",
            "nl"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-ru

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/fr-ru')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "fr",
            "ru"
        ],
        "id": null,
        "_type": "Translation"
    }
}

fr-zh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/fr-zh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "fr",
            "zh"
        ],
        "id": null,
        "_type": "Translation"
    }
}

nl-ru

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/nl-ru')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "nl",
            "ru"
        ],
        "id": null,
        "_type": "Translation"
    }
}

nl-zh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/nl-zh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "nl",
            "zh"
        ],
        "id": null,
        "_type": "Translation"
    }
}

ru-zh

Utilisez la commande suivante pour charger cet ensemble de données dans TFDS :

ds = tfds.load('huggingface:opus100/ru-zh')
  • Description :
OPUS-100 is English-centric, meaning that all training pairs include English on either the source or target side.
The corpus covers 100 languages (including English).OPUS-100 contains approximately 55M sentence pairs.
Of the 99 language pairs, 44 have 1M sentence pairs of training data, 73 have at least 100k, and 95 have at least 10k.
  • Licence : Aucune licence connue
  • Version : 0.0.0
  • Divisions :
Diviser Exemples
'test' 2000
  • Caractéristiques :
{
    "translation": {
        "languages": [
            "ru",
            "zh"
        ],
        "id": null,
        "_type": "Translation"
    }
}