--- language: - nl license: apache-2.0 tags: - dutch - t5 - t5x - ul2 - seq2seq datasets: - yhavinga/mc4_nl_cleaned - yhavinga/nedd_wiki_news inference: false --- # ul2-base-dutch for Dutch Pretrained T5 model on Dutch using a UL2 (Mixture-of-Denoisers) objective. The T5 model was introduced in [this paper](https://arxiv.org/abs/1910.10683) and first released at [this page](https://github.com/google-research/text-to-text-transfer-transformer). The UL2 objective was introduced in [this paper](https://arxiv.org/abs/2205.05131) and first released at [this page](https://github.com/google-research/google-research/tree/master/ul2). **Note:** The Hugging Face inference widget is deactivated because this model needs a text-to-text fine-tuning on a specific downstream task to be useful in practice. ## Model description T5 is an encoder-decoder model and treats all NLP problems in a text-to-text format. `ul2-base-dutch` T5 is a transformers model pretrained on a very large corpus of Dutch data in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and outputs from those texts. This model used the [T5 v1.1](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) improvements compared to the original T5 model during the pretraining: - GEGLU activation in the feed-forward hidden layer, rather than ReLU - see [here](https://arxiv.org/abs/2002.05202) - Dropout was turned off during pre-training. Dropout should be re-enabled during fine-tuning - Pre-trained on self-supervised objective only without mixing in the downstream tasks - No parameter sharing between embedding and classifier layer ### UL2 pretraining objective This model was pretrained with the UL2's Mixture-of-Denoisers (MoD) objective, that combines diverse pre-training paradigms together. UL2 frames different objective functions for training language models as denoising tasks, where the model has to recover missing sub-sequences of a given input. During pre-training it uses a novel mixture-of-denoisers that samples from a varied set of such objectives, each with different configurations. UL2 is trained using a mixture of three denoising tasks: 1. R-denoising (or regular span corruption), which emulates the standard T5 span corruption objective; 2. X-denoising (or extreme span corruption); and 3. S-denoising (or sequential PrefixLM). During pre-training, we sample from the available denoising tasks based on user-specified ratios. UL2 introduces a notion of mode switching, wherein downstream fine-tuning is associated with specific pre-training denoising task. During the pre-training, a paradigm token is inserted to the input (`[NLU]` for R-denoising, `[NLG]` for X-denoising, or `[S2S]` for S-denoising) indicating the denoising task at hand. Then, during fine-tuning the same input token should be inserted to get the best performance for different downstream fine-tuning tasks. ## Intended uses & limitations This model was only pretrained in a self-supervised way excluding any supervised training. Therefore, this model has to be fine-tuned before it is usable on a downstream task, like text classification, unlike the Google's original T5 model. **Note:** You most likely need to fine-tune these T5/UL2 models without mixed precision so fine-tune them with full fp32 precision. Fine-tuning with Flax in bf16 - `model.to_bf16()` - is possible if you set the mask correctly to exclude layernorm and embedding layers. Also note that the T5x pre-training and fine-tuning configs set `z_loss` to 1e-4, which is used to keep the loss scale from underflowing. You can also find more fine-tuning tips from [here](https://discuss.huggingface.co/t/t5-finetuning-tips), for example. **Note**: For fine-tuning, most likely you can get better results if you insert a prefix token of `[NLU]`, `[NLG]`, or `[S2S]` to your input texts. For general language understanding fine-tuning tasks, you could use the `[NLU]` token. For GPT-style causal language generation, you could use the `[S2S]` token. The token `[NLG]` of the X-denoising pretrain task is somewhat mix between the language understanding and causal language generation so the token `[NLG]` could maybe be used for language generation fine-tuning too. ### How to use Here is how to use this model in PyTorch: ```python from transformers import T5Tokenizer, T5ForConditionalGeneration tokenizer = T5Tokenizer.from_pretrained("yhavinga/ul2-base-dutch", use_fast=False) model = T5ForConditionalGeneration.from_pretrained("yhavinga/ul2-base-dutch") ``` and in Flax: ```python from transformers import T5Tokenizer, FlaxT5ForConditionalGeneration tokenizer = T5Tokenizer.from_pretrained("yhavinga/ul2-base-dutch", use_fast=False) model = FlaxT5ForConditionalGeneration.from_pretrained("yhavinga/ul2-base-dutch") ``` ### Limitations and bias The training data used for this model contains a lot of unfiltered content from the internet, which is far from neutral. Therefore, the model can have biased predictions. This bias will also affect all fine-tuned versions of this model. ## Training data The `ul2-base-dutch` T5 model was pre-trained simultaneously on a combination of several datasets, including the full version of the "mc4_nl_cleaned" dataset, which is a cleaned version of Common Crawl's web crawl corpus, Dutch books, the Dutch subset of Wikipedia (2022-03-20), and a subset of "mc4_nl_cleaned" containing only texts from Dutch newspapers. ## Training procedure ### Preprocessing The ul2-base-dutch T5 model uses a SentencePiece unigram tokenizer with a vocabulary of 32,000 tokens. The tokenizer includes the special tokens ``, ``, ``, known from the original T5 paper, `[NLU]`, `[NLG]` and `[S2S]` for the MoD pre-training, and `` for newline. During pre-training with the UL2 objective, input and output sequences consist of 512 consecutive tokens. The tokenizer does not lowercase texts and is therefore case-sensitive; it distinguises between `dutch` and `Dutch`. Additionally, 100+28 extra tokens were added for pre-training tasks, resulting in a total of 32,128 tokens. ### Pretraining The model was trained on TPUv3-8 VM, sponsored by the [Google TPU Research Cloud](https://sites.research.google/trc/about/), for 1000000 steps with a batch size of 128 (in total 65 B tokens). The optimizer used was AdaFactor with learning rate warmup for 10K steps with a constant learning rate of 1e-2, and then an inverse square root decay (exponential decay) of the learning rate after. The model was trained with Google's Jax/Flax based [t5x framework](https://github.com/google-research/t5x) with help from [Stephenn Fernandes](https://huggingface.co/StephennFernandes) to get started writing task definitions that wrap HF datasets. The UL2 training objective code used with the [t5x framework](https://github.com/google-research/t5x) was copied and slightly modified from the [UL2 paper](https://arxiv.org/pdf/2205.05131.pdf) appendix chapter 9.2 by the authors of the Finnish ul2 models. Used UL2 objective code is available in the repository [Finnish-NLP/ul2-base-nl36-finnish](https://huggingface.co/Finnish-NLP/ul2-base-nl36-finnish) in the files `ul2_objective.py` and `tasks.py`. UL2's mixture-of-denoisers configuration was otherwise equal to the UL2 paper but for the rate of mixing denoisers, 20% for S-denoising was used (suggested at the paper chapter 4.5) and the rest was divided equally between the R-denoising and X-denoising (i.e. 40% for both). ### Model list Models in this series: | | ul2-base-dutch | ul2-base-nl36-dutch | ul2-large-dutch | ul2-small-dutch | |:---------------------|:---------------------|:----------------------|:---------------------|:---------------------| | model_type | t5 | t5 | t5 | t5 | | _pipeline_tag | text2text-generation | text2text-generation | text2text-generation | text2text-generation | | d_model | 768 | 768 | 1024 | 512 | | d_ff | 2048 | 3072 | 2816 | 1024 | | num_heads | 12 | 12 | 16 | 6 | | d_kv | 64 | 64 | 64 | 64 | | num_layers | 12 | 36 | 24 | 8 | | num_decoder_layers | 12 | 36 | 24 | 8 | | feed_forward_proj | gated-gelu | gated-gelu | gated-gelu | gated-gelu | | dense_act_fn | gelu_new | gelu_new | gelu_new | gelu_new | | vocab_size | 32128 | 32128 | 32128 | 32128 | | tie_word_embeddings | 0 | 0 | 0 | 0 | | torch_dtype | float32 | float32 | float32 | float32 | | _gin_batch_size | 128 | 64 | 64 | 128 | | _gin_z_loss | 0.0001 | 0.0001 | 0.0001 | 0.0001 | | _gin_t5_config_dtype | 'bfloat16' | 'bfloat16' | 'bfloat16' | 'bfloat16' | ## Evaluation results See the evaluation section in the interactive [Pre-training Dutch T5 Models](https://huggingface.co/spaces/yhavinga/pre-training-dutch-t5-models) blog. ## Acknowledgements This project would not have been possible without compute generously provided by Google through the [TPU Research Cloud](https://sites.research.google/trc/). Thanks to the [Finnish-NLP](https://huggingface.co/Finnish-NLP) authors for releasing their code for the UL2 objective and associated task definitions. Thanks to [Stephenn Fernandes](https://huggingface.co/StephennFernandes) for helping me get started with the t5x framework. Created by [Yeb Havinga](https://www.linkedin.com/in/yeb-havinga-86530825/)