Add mlm pretraining xla torch readme (#12011)

* fix_torch_device_generate_test

* remove @

* upload

* Apply suggestions from code review

* Apply suggestions from code review

* Apply suggestions from code review

* Update examples/flax/language-modeling/README.md

* add more info

* finish

* fix

Co-authored-by: Patrick von Platen <patrick@huggingface.co>
This commit is contained in:
Patrick von Platen 2021-06-14 10:31:21 +01:00 committed by GitHub
parent ecd6efe7cb
commit 16c0efca2c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 119 additions and 3 deletions

View File

@ -123,7 +123,123 @@ This should take less than 18 hours.
Training statistics can be accessed on [tfhub.de](https://tensorboard.dev/experiment/GdYmdak2TWeVz0DDRYOrrg).
For a step-by-step walkthrough of how to do masked language modeling in Flax, please have a
look at [this TODO: (Patrick)]() google colab.
look at [this](https://colab.research.google.com/github/huggingface/notebooks/blob/master/examples/masked_language_modeling_flax.ipynb) google colab.
## TODO(Patrick): Add comparison with PyTorch GPU/TPU
## Runtime evaluation
We also ran masked language modeling using PyTorch/XLA on a TPUv3-8, and PyTorch on 8 V100 GPUs. We report the
overall training time below.
For reproducibility, we state the training commands used for PyTorch/XLA and PyTorch further below.
| Task | [TPU v3-8 (Flax)](https://tensorboard.dev/experiment/GdYmdak2TWeVz0DDRYOrrg/) | [TPU v3-8 (Pytorch/XLA)](https://tensorboard.dev/experiment/7Jq1kcQQRAmy12KOdXek7A/)| [8 GPU (PyTorch)](https://tensorboard.dev/experiment/PJneV8FQRxa2unPw1QnVHA) |
|-------|-----------|------------|------------|
| MLM | 15h32m | 23h46m | 44h14m |
| **COST*** | $124.24 | $187.84 | $877.92 |
*All experiments are ran on Google Cloud Platform. Prices are on-demand prices
(not preemptible), obtained on May 12, 2021 for zone Iowa (us-central1) using
the following tables:
[TPU pricing table](https://cloud.google.com/tpu/pricing) ($8.00/h for v3-8),
[GPU pricing table](https://cloud.google.com/compute/gpus-pricing) ($2.48/h per
V100 GPU). GPU experiments are ran without further optimizations besides JAX
transformations. GPU experiments are ran with full precision (fp32). "TPU v3-8"
are 8 TPU cores on 4 chips (each chips has 2 cores), while "8 GPU" are 8 GPU chips.
### Script to run MLM with PyTorch/XLA on TPUv3-8
For comparison one can run the same pre-training with PyTorch/XLA on TPU. To set up PyTorch/XLA on Cloud TPU VMs, please
refer to [this](https://cloud.google.com/tpu/docs/pytorch-xla-ug-tpu-vm) guide.
Having created the tokenzier and configuration in `norwegian-roberta-base`, we create the following symbolic links:
```bash
ln -s ~/transformers/examples/pytorch/language-modeling/run_mlm.py ./
ln -s ~/transformers/examples/pytorch/xla_spawn.py ./
```
, set the following environment variables:
```bash
export XRT_TPU_CONFIG="localservice;0;localhost:51011"
unset LD_PRELOAD
export NUM_TPUS=8
export TOKENIZERS_PARALLELISM=0
export MODEL_DIR="./norwegian-roberta-base"
mkdir -p ${MODEL_DIR}
```
, and start training as follows:
```bash
python3 xla_spawn.py --num_cores ${NUM_TPUS} run_mlm.py --output_dir="./runs" \
--model_type="roberta" \
--config_name="${MODEL_DIR}" \
--tokenizer_name="${MODEL_DIR}" \
--dataset_name="oscar" \
--dataset_config_name="unshuffled_deduplicated_no" \
--max_seq_length="128" \
--weight_decay="0.01" \
--per_device_train_batch_size="128" \
--per_device_eval_batch_size="128" \
--learning_rate="3e-4" \
--warmup_steps="1000" \
--overwrite_output_dir \
--num_train_epochs="18" \
--adam_beta1="0.9" \
--adam_beta2="0.98" \
--do_train \
--do_eval \
--logging_steps="500" \
--evaluation_strategy="epoch" \
--report_to="tensorboard" \
--save_strategy="no"
```
### Script to compare pre-training with PyTorch on 8 GPU V100's
For comparison you can run the same pre-training with PyTorch on GPU. Note that we have to make use of `gradient_accumulation`
because the maximum batch size that fits on a single V100 GPU is 32 instead of 128.
Having created the tokenzier and configuration in `norwegian-roberta-base`, we create the following symbolic links:
```bash
ln -s ~/transformers/examples/pytorch/language-modeling/run_mlm.py ./
```
, set some environment variables:
```bash
export NUM_GPUS=8
export TOKENIZERS_PARALLELISM=0
export MODEL_DIR="./norwegian-roberta-base"
mkdir -p ${MODEL_DIR}
```
, and can start training as follows:
```bash
python3 -m torch.distributed.launch --nproc_per_node ${NUM_GPUS} run_mlm.py \
--output_dir="./runs" \
--model_type="roberta" \
--config_name="${MODEL_DIR}" \
--tokenizer_name="${MODEL_DIR}" \
--dataset_name="oscar" \
--dataset_config_name="unshuffled_deduplicated_no" \
--max_seq_length="128" \
--weight_decay="0.01" \
--per_device_train_batch_size="32" \
--per_device_eval_batch_size="32" \
--gradient_accumulation="4" \
--learning_rate="3e-4" \
--warmup_steps="1000" \
--overwrite_output_dir \
--num_train_epochs="18" \
--adam_beta1="0.9" \
--adam_beta2="0.98" \
--do_train \
--do_eval \
--logging_steps="500" \
--evaluation_strategy="steps" \
--report_to="tensorboard" \
--save_strategy="no"
```

View File

@ -101,7 +101,7 @@ overall training time below. For comparison we ran Pytorch's [run_glue.py](https
*All experiments are ran on Google Cloud Platform. Prices are on-demand prices
(not preemptible), obtained on May 12, 2021 for zone Iowa (us-central1) using
the following tables:
[TPU pricing table](https://cloud.google.com/tpu/pricing) ($2.40/h for v3-8),
[TPU pricing table](https://cloud.google.com/tpu/pricing) ($8.00/h for v3-8),
[GPU pricing table](https://cloud.google.com/compute/gpus-pricing) ($2.48/h per
V100 GPU). GPU experiments are ran without further optimizations besides JAX
transformations. GPU experiments are ran with full precision (fp32). "TPU v3-8"