File size: 10,564 Bytes
2e3422b 99ca5a1 2e3422b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 |
---
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 `<pad>`, `</s>`, `<unk>`, known from the original T5 paper,
`[NLU]`, `[NLG]` and `[S2S]` for the MoD pre-training, and `<n>` 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/)
|