Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/zphang/architecture-objective


https://github.com/zphang/architecture-objective

Last synced: 3 months ago
JSON representation

Awesome Lists containing this project

README 

        
# What Language Model Architecture and Pretraining Objective Work Best for Zero-Shot Generalization?



Large pretrained Transformer language models have been shown to exhibit zero-shot generalization, i.e. they can perform a wide variety of tasks that they were not explicitly trained on. However, the architectures and pretraining objectives used across state-of-the-art models differ significantly, and there has been limited systematic comparison of these factors. In this work, we present a large-scale evaluation of modeling choices and their impact on zero-shot generalization. In particular, we focus on text-to-text models and experiment with three model architectures (causal/non-causal decoder-only and encoder-decoder), trained with two different pretraining objectives (autoregressive and masked language modeling), and evaluated with and without multitask prompted finetuning. We train models with over 5 billion parameters for more than 170 billion tokens, thereby increasing the likelihood that our conclusions will transfer to even larger scales. Our experiments show that causal decoder-only models trained on an autoregressive language modeling objective exhibit the strongest zero-shot generalization after purely unsupervised pretraining. However, models with non-causal visibility on their input trained with a masked language modeling objective followed by multitask finetuning perform the best among our experiments. We therefore consider the adaptation of pretrained models across architectures and objectives. We find that pretrained non-causal decoder models can be adapted into performant generative causal decoder models, using autoregressive language modeling as a downstream task. Furthermore, we find that pretrained causal decoder models can be efficiently adapted into non-causal decoder models, ultimately achieving competitive performance after multitask finetuning. 



Full paper is available at: https://arxiv.org/abs/2204.05832



## Checkpoints

// TODO (@thomasw21):  link to checkpoints.



## How to cite



    @article{wang2022language,

      title={What Language Model Architecture and Pretraining Objective Work Best for Zero-Shot Generalization?},

      author={Wang, Thomas and Roberts, Adam and Hesslow, Daniel and Scao, Teven Le and Chung, Hyung Won and Beltagy, Iz and Launay, Julien and Raffel, Colin},

      journal={arXiv preprint arXiv:2204.05832},

      year={2022}

    }



# T5X



T5X is a modular, composable, research-friendly framework for high-performance,

configurable, self-service training, evaluation, and inference of sequence

models (starting with language) at many scales.



It is essentially a new and improved implementation of the

[T5 codebase](https://github.com/google-research/text-to-text-transfer-transformer)

(based on [Mesh TensorFlow](https://github.com/tensorflow/mesh)) in [JAX](https://github.com/google/jax) and [Flax](https://github.com/google/flax).



## Installation



Note that all the commands in this document should be run in the commandline of

the TPU VM instance unless otherwise stated.



1.  Follow the

    [instructions](https://cloud.google.com/tpu/docs/jax-quickstart-tpu-vm#install_the_google_cloud_sdk)

    to set up a Google Cloud Platform (GCP) account and enable the Cloud TPU

    API.



    **Note:** While T5X works with GPU as well, we haven't heavily tested the

    GPU usage.



2.  Create a

    [Cloud TPU VM instance](https://cloud.google.com/blog/products/compute/introducing-cloud-tpu-vms)

    following

    [this instruction](https://cloud.google.com/tpu/docs/jax-quickstart-tpu-vm#create-vm).

    We recommend that you develop your workflow in a single v3-8 TPU (i.e.,

    `--accelerator-type=v3-8`) and scale up to pod slices once the pipeline is

    ready. In this README, we focus on using a single v3-8 TPU. See

    [here](https://cloud.google.com/tpu/docs/system-architecture-tpu-vm) to

    learn more about TPU architectures.



3.  With Cloud TPU VMs, you ssh directly into the host machine of the TPU VM.

    You can install packages, run your code run, etc. in the host machine. Once

    the TPU instance is created, ssh into it with



    ```sh

    gcloud alpha compute tpus tpu-vm ssh ${TPU_NAME} --zone=${ZONE}

    ```



    where `TPU_NAME` and `ZONE` are the name and the zone used in step 2.



4.  Install T5X and the dependencies. JAX and Gin-config need to be installed

    from the source.



    ```sh

    git clone --branch=main https://github.com/google-research/t5x

    cd t5x



    python3 -m pip install -e . -f \

      https://storage.googleapis.com/jax-releases/libtpu_releases.html



    ```



5.  Create toogle Cloud Storage (GCS) bucket to store the dataset and model

    checkpoints. To create a GCS bucket, see these

    [instructions](https://cloud.google.com/storage/docs/creating-buckets).



## Example: English to German translation



As a running example, we use the WMT14 En-De translation. The raw dataset is

available in TensorFlow Datasets as

["wmt_t2t_translate"](https://www.tensorflow.org/datasets/catalog/wmt_t2t_translate).



T5 casts the translation task such as the following



```py

{'en': 'That is good.', 'de': 'Das ist gut.'}

```



to the form called "text-to-text":



```py

{'inputs': 'translate English to German: That is good.', 'targets': 'Das ist gut.'}

```



This formulation allows many different classes of language tasks to be expressed

in a uniform manner and a single encoder-decoder architecture can handle them

without any task-specific parameters. For more detail, refer to the [T5 paper

(Raffel et al. 2019)][t5_paper].



For a scalable data pipeline and an evaluation framework, we use

[`SeqIO`](https://github.com/google/seqio), which was factored out of the [T5

library][t5_github]. A `seqio.Task` packages together the raw dataset, vocabulary,

preprocessing such as tokenization and evaluation metrics such as

[BLEU](https://aclanthology.org/P02-1040.pdf) and provides a

[`tf.data`](https://www.tensorflow.org/guide/data) instance.



[The T5 library][t5_github] provides a number of `seqio.Task`s that were used in the

[T5 paper][t5_paper]. In this example, we use [wmt_t2t_ende_v003](https://github.com/google-research/text-to-text-transfer-transformer/blob/d81c0bab2a41b4d5dfbe4971de32f7d67df65f31/t5/data/tasks.py#L212).



### Training



To run a training job, we use the `t5x/train.py` script.



```sh

# Model dir to save logs, ckpts, etc. in "gs://model_dir" format.

MODEL_DIR="..."



# Data dir to save the processed dataset in "gs://data_dir" format.

TFDS_DATA_DIR="..."

T5X_DIR="..."  # directory where the T5X repo is cloned.



python3 ${T5X_DIR}/t5x/train.py \

  --gin_file="t5x/examples/t5/t5_1_1/examples/t5_1_1_base_wmt_from_scratch.gin" \

  --gin.MODEL_DIR="'${MODEL_DIR}'" \

  --tfds_data_dir=${TFDS_DATA_DIR}

```



The configuration for this training run is defined in the Gin file

[t5_1_1_base_wmt_from_scratch.gin](t5x/examples/t5/t5_1_1/examples/t5_1_1_base_wmt_from_scratch.gin).

[Gin-config](https://github.com/google/gin-config) is a library to handle

configurations based on dependency injection. Among many benefits, Gin allows

users to pass custom components such as a custom model to the T5X library

without having to modify the core library. The [custom

components](#custom-components) section shows how this is done.



While the core library is independent of Gin, it is central to the examples we

provide. Therefore, we provide a short [introduction][gin-primer] to Gin in the

context of T5X.  All the configurations are written to a file "config.gin" in

`MODEL_DIR`. This makes debugging as well as reproducing the experiment much

easier.



In addition to the `config.json`, `model-info.txt` file summarizes the model

parameters (shape, names of the axes, partitioning info) as well as the

optimizer states.



#### TensorBoard



To monitor the training in [TensorBoard](https://www.tensorflow.org/tensorboard), it is much easier (due to

authentification issues) to launch the TensorBoard on your own machine and _not_ in

the TPU VM. So in the commandline where you ssh'ed into the TPU VM, launch the

TensorBoard with the `logdir` pointing to the `MODEL_DIR`.



```sh

# NB: run this on your machine not TPU VM!

MODEL_DIR="..."  # Copy from the TPU VM.

tensorboard --logdir=${MODEL_DIR}

```



Or you can launch the TensorBoard inside a Colab. In a Colab cell, run



```python

from google.colab import auth

auth.authenticate_user()

```



to authorize the Colab to access the GCS bucket and launch the TensorBoard.



```python

%load_ext tensorboard

model_dir = "..."  # Copy from the TPU VM.

%tensorboard --logdir=model_dir

```



TODO(hwchung): Add tfds preparation instruction



### Fine-tuning



We can leverage the benefits of self-supervised pre-training by initializing

from one of our pre-trained models. Here we use the T5.1.1 Base checkpoint.



```sh

# Model dir to save logs, ckpts, etc. in "gs://model_dir" format.

MODEL_DIR="..."



# Data dir to save the processed dataset in "gs://data_dir" format.

TFDS_DATA_DIR="..."

T5X_DIR="..."  # directory where the T5X repo is cloned.



python3 ${T5X_DIR}/t5x/train.py \

  --gin_file="t5x/examples/t5/t5_1_1/examples/t5_1_1_base_wmt_finetune.gin" \

  --gin.MODEL_DIR="'${MODEL_DIR}'" \

  --tfds_data_dir=${TFDS_DATA_DIR}

```



**Note:** when supplying a string, dict, list, tuple value, or a bash variable

via a flag, you must put it in quotes. In the case of strings, it requires

"triple quotes" (`"''"`). For example:

`--gin.utils.DatasetConfig.split="'validation'"` or

`--gin.MODEL_DIR="'${MODEL_DIR}'"`.



Gin makes it easy to change a number of configurations. For example, you can

change the `partitioning.ModelBasedPjitPartitioner.num_partitions` (overriding

the value in

[t5_1_1_base_wmt_from_scratch.gin](t5x/examples/t5/t5_1_1/examples/t5_1_1_base_wmt_from_scratch.gin))

to chanage the parallelism strategy and pass it as a commandline arg.



```sh

--gin.partitioning.ModelBasedPjitPartitioner.num_partitions=8

```



### Evaluation



To run the offline (i.e. without training) evaluation, you can use `t5x/eval.py`

script.



```sh

EVAL_OUTPUT_DIR="..."  # directory to write eval output

T5X_DIR="..."  # directory where the t5x is cloned, e.g., ${HOME}"/t5x".

TFDS_DATA_DIR="..."

CHECKPOINT_PATH="..."



python3 ${T5X_DIR}/t5x/eval.py \

  --gin_file="t5x/examples/t5/t5_1_1/examples/t5_1_1_base_wmt_eval.gin" \

  --gin.CHECKPOINT_PATH="'${CHECKPOINT_PATH}'" \

  --gin.EVAL_OUTPUT_DIR="'${EVAL_OUTPUT_DIR}'" \

  --tfds_data_dir=${TFDS_DATA_DIR}

```



### Inference



To run inference, you can use `t5x/infer.py` script. Here we use the same

`seqio.Task`, but for inference we do not use the targets features other than

logging them alongside the prediction in a JSON file.



```sh

INFER_OUTPUT_DIR="..."  # directory to write infer output

T5X_DIR="..."  # directory where the t5x is cloned, e.g., ${HOME}"/t5x".

TFDS_DATA_DIR="..."

CHECKPOINT_PATH="..."



python3 ${T5X_DIR}/t5x/infer.py \

  --gin_file="t5x/examples/t5/t5_1_1/examples/t5_1_1_base_wmt_infer.gin" \

  --gin.CHECKPOINT_PATH="'${CHECKPOINT_PATH}'" \

  --gin.INFER_OUTPUT_DIR="'${INFER_OUTPUT_DIR}'" \

  --tfds_data_dir=${TFDS_DATA_DIR}

```



## Custom components



[The translation example](#example-english-to-german-translation) uses the

encoder-decoder model that T5X provides as well as the dataset from the T5

library. This section shows how you can use your own dataset and a model and

pass via Gin.



### Example: custom dataset in a user directory



For this example, we have the following directory structure with

`${HOME}/dir1/user_dir` representing a user directory with custom components.



```

${HOME}

└── dir1

    └── user_dir

        ├── t5_1_1_base_de_en.gin

        └── tasks.py

```



As an example, let's define a new dataset. Here we use the same Translation

dataset but we define the translation task in the opposite direction, i.e.,

German to English intead of English to German. We define this task in `tasks.py`



```py

# ${HOME}/dir1/user_dir/tasks.py



import functools

import seqio

import tensorflow_datasets as tfds

from t5.evaluation import metrics

from t5.data import preprocessors



vocabulary = seqio.SentencePieceVocabulary(

    'gs://t5-data/vocabs/cc_all.32000/sentencepiece.model', extra_ids=100)

output_features = {

    'inputs': seqio.Feature(vocabulary=vocabulary),

    'targets': seqio.Feature(vocabulary=vocabulary)

}



seqio.TaskRegistry.add(

    'wmt_t2t_de_en_v003',

    source=seqio.TfdsDataSource(tfds_name='wmt_t2t_translate/de-en:1.0.0'),

    preprocessors=[

        functools.partial(

            preprocessors.translate,

            source_language='de', target_language='en'),

        seqio.preprocessors.tokenize,

        seqio.CacheDatasetPlaceholder(),

        seqio.preprocessors.append_eos_after_trim,

    ],

    metric_fns=[metrics.bleu],

    output_features=output_features)

```



In the Gin file, most of the settings are equivalent to those used in the

[En->De example](#example-english-to-german-translation). So we include the Gin

file from that example. To use "wmt_t2t_de_en_v003" task we just defined, we

need to import the task module "tasks.py". Note that we use a relative path

defined with respect to the user directory. This will be specified as a

flag.



```py

# ${HOME}/dir1/user_dir/t5_1_1_base_de_en.gin

from __gin__ import dynamic_registration

import tasks  # This imports the task defined in dir1/user_dir/tasks.py.



include "t5x-tmp/t5x/examples/t5/t5_1_1/examples/t5_1_1_base_wmt_from_scratch.gin"

MIXTURE_OR_TASK_NAME = "wmt_t2t_de_en_v003"

```



Finally, we launch training passing the user directory as a flag

`gin_search_paths` such that the Gin file and python modules can be specified

with relative paths.



```sh

PROJECT_DIR=${HOME}"/dir1/user_dir"

T5X_DIR="..."  # directory where the t5x is cloned.

TFDS_DATA_DIR="..."

MODEL_DIR="..."

export PYTHONPATH=${PROJECT_DIR}



python3 ${T5X_DIR}/t5x/train.py \

  --gin_search_paths=${PROJECT_DIR} \

  --gin_file="t5_1_1_base_de_en.gin" \

  --gin.MODEL_DIR="'${MODEL_DIR}'" \

  --tfds_data_dir=${TFDS_DATA_DIR}

```



## Released Checkpoints



We release the checkpoints for the T5.1.1 models in a native T5X format.



* **t5.1.1.small** (~77 million parameters): [gs://t5-data/pretrained_models/t5x/t5_1_1_small/checkpoint_1000000](https://console.cloud.google.com/storage/browser/t5-data/pretrained_models/t5x/t5_1_1_small/checkpoint_1000000)

* **t5.1.1.base** (~250 million parameters): [gs://t5-data/pretrained_models/t5x/t5_1_1_base/checkpoint_1000000](https://console.cloud.google.com/storage/browser/t5-data/pretrained_models/t5x/t5_1_1_base/checkpoint_1000000)

* **t5.1.1.large** (~800 million parameters): [gs://t5-data/pretrained_models/t5x/t5_1_1_large/checkpoint_1000000](https://console.cloud.google.com/storage/browser/t5-data/pretrained_models/t5x/t5_1_1_large/checkpoint_1000000)

* **t5.1.1.xl** (~3 billion parameters): [gs://t5-data/pretrained_models/t5x/t5_1_1_xl/checkpoint_1000000](https://console.cloud.google.com/storage/browser/t5-data/pretrained_models/t5x/t5_1_1_xl/checkpoint_1000000)

* **t5.1.1.xxl** (~11 billion parameters): [gs://t5-data/pretrained_models/t5x/t5_1_1_xxl/checkpoint_1000000](https://console.cloud.google.com/storage/browser/t5-data/pretrained_models/t5x/t5_1_1_xxl/checkpoint_1000000)



These are converted from the public [Mesh TensorFlow

checkpoints](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511)

.



## Compatibility with the Mesh TensorFlow checkpoints

The Mesh TensorFlow checkpoints trained using the [T5 library][t5_github] can be

directly loaded into T5X. For example, we can rerun the fine-tuning example

initializing from the MTF checkpoint by changing the `INIT_CHECKPOINT` Gin

macro.



```sh

# Model dir to save logs, ckpts, etc. in "gs://model_dir" format.

MODEL_DIR="..."



# Data dir to save the processed dataset in "gs://data_dir" format.

TFDS_DATA_DIR="..."

T5X_DIR="..."  # directory where the T5X repo is cloned.



python3 ${T5X_DIR}/t5x/train.py \

  --gin_file="t5x/examples/t5/t5_1_1/examples/wmt19_ende_from_scratch.gin" \

  --gin.MODEL_DIR="'${MODEL_DIR}'" \

  --gin.MIXTURE_OR_TASK_NAME="'wmt_t2t_ende_v003'" \

  --gin.INIT_CHECKPOINT="'gs://t5-data/pretrained_models/t5.1.1.base/model.ckpt-1000000'" \

  --tfds_data_dir=${TFDS_DATA_DIR}

```



Note that restoring directly from the Mesh TensorFlow checkpoints can be

inefficient if heavy model parallelism is used for large models. This is

because each host loads the entire copy of the model first and then keep only

the relevant slices dictated by the model parallelism specification. If you have

Mesh TensorFlow checkpoints that you run often, we recommend converting the

checkpoints to T5X native format using

[`Checkpointer.convert_from_tf_checkpoint`](https://github.com/google-research/t5x/blob/fba685d1d49bfb1000f37b5952a9a0533f24ed36/t5x/checkpoints.py#L886).



TODO(hwchung): Add a conversion script.



## Note

This is not an officially supported Google product



[t5_paper]: https://arxiv.org/abs/1910.10683

[t5_github]: https://github.com/google-research/text-to-text-transfer-transformer

[gin-primer]: gin-primer.md