--- datasets: - eduagarcia/CrawlPT_dedup language: - pt pipeline_tag: fill-mask model-index: - name: RoBERTaCrawlPT-base results: - task: type: token-classification dataset: type: lener_br name: lener_br split: test metrics: - type: seqeval value: 0.8924 name: F1 args: scheme: IOB2 - task: type: token-classification dataset: type: eduagarcia/PortuLex_benchmark name: UlyNER-PL Coarse config: UlyssesNER-Br-PL-coarse split: test metrics: - type: seqeval value: 0.8822 name: F1 args: scheme: IOB2 - task: type: token-classification dataset: type: eduagarcia/PortuLex_benchmark name: UlyNER-PL Fine config: UlyssesNER-Br-PL-fine split: test metrics: - type: seqeval value: 0.8658 name: F1 args: scheme: IOB2 - task: type: token-classification dataset: type: eduagarcia/PortuLex_benchmark name: FGV-STF config: fgv-coarse split: test metrics: - type: seqeval value: 0.7988 name: F1 args: scheme: IOB2 - task: type: token-classification dataset: type: eduagarcia/PortuLex_benchmark name: RRIP config: rrip split: test metrics: - type: seqeval value: 0.8280 name: F1 args: scheme: IOB2 - task: type: token-classification dataset: type: eduagarcia/PortuLex_benchmark name: PortuLex split: test metrics: - type: seqeval value: 0.8483 name: Average F1 args: scheme: IOB2 license: cc-by-4.0 metrics: - seqeval --- # RoBERTaCrawlPT-base RoBERTaCrawlPT-base is a generic Portuguese Masked Language Model pretrained from scratch from the [CrawlPT](https://huggingface.co/datasets/eduagarcia/CrawlPT_dedup) corpora, using the same architecture as [RoBERTa-base](https://huggingface.co/FacebookAI/roberta-base). This model is part of the [RoBERTaLexPT](https://huggingface.co/eduagarcia/RoBERTaLegalPT-base) work: [Coming soon] - **Language(s) (NLP):** Brazilian Portuguese (pt-BR) - **License:** [Creative Commons Attribution 4.0 International Public License](https://creativecommons.org/licenses/by/4.0/deed.en) - **Repository:** https://github.com/eduagarcia/roberta-legal-portuguese - **Paper:** [Coming soon] ## Generic Evaluation TO-DO... ## Legal Evaluation The model was evaluated on ["PortuLex" benchmark](https://huggingface.co/datasets/eduagarcia/PortuLex_benchmark), a four-task benchmark designed to evaluate the quality and performance of language models in the Portuguese legal domain. Macro F1-Score (\%) for multiple models evaluated on PortuLex benchmark test splits: | **Model** | **LeNER** | **UlyNER-PL** | **FGV-STF** | **RRIP** | **Average (%)** | |----------------------------------------------------------------------------|-----------|-----------------|-------------|:---------:|-----------------| | | | Coarse/Fine | Coarse | | | | [BERTimbau-base](https://huggingface.co/neuralmind/bert-base-portuguese-cased) | 88.34 | 86.39/83.83 | 79.34 | 82.34 | 83.78 | | [BERTimbau-large](https://huggingface.co/neuralmind/bert-large-portuguese-cased) | 88.64 | 87.77/84.74 | 79.71 | **83.79** | 84.60 | | [Albertina-PT-BR-base](https://huggingface.co/PORTULAN/albertina-ptbr-based) | 89.26 | 86.35/84.63 | 79.30 | 81.16 | 83.80 | | [Albertina-PT-BR-xlarge](https://huggingface.co/PORTULAN/albertina-ptbr) | 90.09 | 88.36/**86.62** | 79.94 | 82.79 | 85.08 | | [BERTikal-base](https://huggingface.co/felipemaiapolo/legalnlp-bert) | 83.68 | 79.21/75.70 | 77.73 | 81.11 | 79.99 | | [JurisBERT-base](https://huggingface.co/alfaneo/jurisbert-base-portuguese-uncased) | 81.74 | 81.67/77.97 | 76.04 | 80.85 | 79.61 | | [BERTimbauLAW-base](https://huggingface.co/alfaneo/bertimbaulaw-base-portuguese-cased) | 84.90 | 87.11/84.42 | 79.78 | 82.35 | 83.20 | | [Legal-XLM-R-base](https://huggingface.co/joelniklaus/legal-xlm-roberta-base) | 87.48 | 83.49/83.16 | 79.79 | 82.35 | 83.24 | | [Legal-XLM-R-large](https://huggingface.co/joelniklaus/legal-xlm-roberta-large) | 88.39 | 84.65/84.55 | 79.36 | 81.66 | 83.50 | | [Legal-RoBERTa-PT-large](https://huggingface.co/joelniklaus/legal-portuguese-roberta-large) | 87.96 | 88.32/84.83 | 79.57 | 81.98 | 84.02 | | **Ours** | | | | | | | RoBERTaTimbau-base (Reproduction of BERTimbau) | 89.68 | 87.53/85.74 | 78.82 | 82.03 | 84.29 | | RoBERTaLegalPT-base (Trained on LegalPT) | 90.59 | 85.45/84.40 | 79.92 | 82.84 | 84.57 | | **RoBERTaCrawlPT-base (this)** (Trained on CrawlPT) | 89.24 | 88.22/86.58 | 79.88 | 82.80 | 84.83 | | [RoBERTaLexPT-base](https://huggingface.co/eduagarcia/RoBERTaLexPT-base) (Trained on CrawlPT + LegalPT) | **90.73** | **88.56**/86.03 | **80.40** | 83.22 | **85.41** | ## Training Details RoBERTaCrawlPT is pretrained on: - [CrawlPT](https://huggingface.co/datasets/eduagarcia/CrawlPT_dedup) is a composition of three Portuguese general corpora: [brWaC](https://huggingface.co/datasets/brwac), [CC100 PT subset](https://huggingface.co/datasets/eduagarcia/cc100-pt), [OSCAR-2301 PT subset](https://huggingface.co/datasets/eduagarcia/OSCAR-2301-pt_dedup). ### Training Procedure Our pretraining process was executed using the [Fairseq library v0.10.2](https://github.com/facebookresearch/fairseq/tree/v0.10.2) on a DGX-A100 cluster, utilizing a total of 2 Nvidia A100 80 GB GPUs. The complete training of a single configuration takes approximately three days. This computational cost is similar to the work of [BERTimbau-base](https://huggingface.co/neuralmind/bert-base-portuguese-cased), exposing the model to approximately 65 billion tokens during training. #### Preprocessing We deduplicated all subsets of the CrawlPT Corpus using the a MinHash algorithm and Locality Sensitive Hashing implementation from the libary [text-dedup](https://github.com/ChenghaoMou/text-dedup) to find clusters of duplicate documents. To ensure that domain models are not constrained by a generic vocabulary, we utilized the [HuggingFace Tokenizers](https://github.com/huggingface/tokenizers) -- BPE algorithm to train a vocabulary for each pre-training corpus used. #### Training Hyperparameters The pretraining process involved training the model for 62,500 steps, with a batch size of 2048 and a learning rate of 4e-4, each sequence containing a maximum of 512 tokens. The weight initialization is random. We employed the masked language modeling objective, where 15\% of the input tokens were randomly masked. The optimization was performed using the AdamW optimizer with a linear warmup and a linear decay learning rate schedule. For other parameters we adopted the standard [RoBERTa-base hyperparameters](https://huggingface.co/FacebookAI/roberta-base): | **Hyperparameter** | **RoBERTa-base** | |------------------------|-----------------:| | Number of layers | 12 | | Hidden size | 768 | | FFN inner hidden size | 3072 | | Attention heads | 12 | | Attention head size | 64 | | Dropout | 0.1 | | Attention dropout | 0.1 | | Warmup steps | 6k | | Peak learning rate | 4e-4 | | Batch size | 2048 | | Weight decay | 0.01 | | Maximum training steps | 62.5k | | Learning rate decay | Linear | | AdamW $$\epsilon$$ | 1e-6 | | AdamW $$\beta_1$$ | 0.9 | | AdamW $$\beta_2$$ | 0.98 | | Gradient clipping | 0.0 | ## Citation ``` @InProceedings{garcia2024_roberlexpt, author="Garcia, Eduardo A. S. and Silva, N{\'a}dia F. F. and Siqueira, Felipe and Gomes, Juliana R. S. and Albuqueruqe, Hidelberg O. and Souza, Ellen and Lima, Eliomar and De Carvalho, André", title="RoBERTaLexPT: A Legal RoBERTa Model pretrained with deduplication for Portuguese", booktitle="Computational Processing of the Portuguese Language", year="2024", publisher="Association for Computational Linguistics" } ``` ## Acknowledgment This work has been supported by the AI Center of Excellence (Centro de Excelência em Inteligência Artificial – CEIA) of the Institute of Informatics at the Federal University of Goiás (INF-UFG).