mbart25-multilingual-summarization-multilarge-cs

This model is a fine-tuned checkpoint of facebook/mbart-large-cc25 on the Multilingual large summarization dataset focused on Czech texts to produce multilingual summaries.

Task

The model deals with a multi-sentence summary in eight different languages. With the idea of adding other foreign language documents, and by having a considerable amount of Czech documents, we aimed to improve model summarization in the Czech language. Supported languages: 'en_XX' : 'en', 'de_DE': 'de', 'es_XX': 'es', 'fr_XX':'fr', 'ru_RU':'ru', 'tr_TR':'tr'.

USAGE

Assume that you are using the provided MultilingualSummarizer.ipynb file and included files from git repository.

## Configuration of summarization pipeline
#
def summ_config():
    cfg = OrderedDict([
        
        ## summarization model - checkpoint
        #   ctu-aic/m2m100-418M-multilingual-summarization-multilarge-cs
        #   ctu-aic/mt5-base-multilingual-summarization-multilarge-cs
        #   ctu-aic/mbart25-multilingual-summarization-multilarge-cs
        ("model_name", "ctu-aic/mbart25-multilingual-summarization-multilarge-cs"),
        
        ## language of summarization task
        #   language : string : cs, en, de, fr, es, tr, ru, zh
        ("language", "en"), 
        
        ## generation method parameters in dictionary
        #
        ("inference_cfg", OrderedDict([
            ("num_beams", 4),
            ("top_k", 40),
            ("top_p", 0.92),
            ("do_sample", True),
            ("temperature", 0.95),
            ("repetition_penalty", 1.23),
            ("no_repeat_ngram_size", None),
            ("early_stopping", True),
            ("max_length", 128),
            ("min_length", 10),
        ])),
        #texts to summarize values = (list of strings, string, dataset)
        ("texts",
            [
               "english text1 to summarize",
               "english text2 to summarize",
            ]
        ),
        #OPTIONAL: Target summaries values = (list of strings, string, None)
        ('golds',
         [
               "target english text1",
               "target english text2",
         ]),
        #('golds', None),
    ])
    return cfg

cfg = summ_config()
msummarizer = MultiSummarizer(**cfg)
ret = msummarizer(**cfg)

Dataset

Multilingual large summarization dataset consists of 10 sub-datasets mainly based on news and daily mails. For the training, it was used the entire training set and 72% of the validation set.

Train set:        3 464 563 docs
Validation set:     121 260 docs

Truncation and padding were set to 512 tokens for the encoder (input text) and 128 for the decoder (summary).

Training

Trained based on cross-entropy loss.

Time: 3 days 8 hours
Epochs: 860K steps cca 8 (from 10)
GPUs: 4x NVIDIA A100-SXM4-40GB
eloss: 2.214 - 1.762
tloss: 3.365 - 1.445

ROUGE results per individual dataset test set: