[AIR][Docs] Set logging_strategy="epoch" for HF (ray-project#27917)

Signed-off-by: Stefan van der Kleij <[email protected]>
Stefan-1313 · Aug 18, 2022 · 952f401 · 952f401
1 parent 53197ac
commit 952f401
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 2 deletions.
diff --git a/doc/source/ray-air/doc_code/hf_trainer.py b/doc/source/ray-air/doc_code/hf_trainer.py
@@ -62,6 +62,8 @@ def trainer_init_per_worker(train_dataset, eval_dataset, **config):
  args = transformers.TrainingArguments(
  output_dir=f"{model_checkpoint}-wikitext2",
  evaluation_strategy="epoch",
+ save_strategy="epoch",
+ logging_strategy="epoch",
  learning_rate=2e-5,
  weight_decay=0.01,
  no_cuda=True, # Set to False for GPU training

diff --git a/doc/source/ray-air/examples/huggingface_text_classification.ipynb b/doc/source/ray-air/examples/huggingface_text_classification.ipynb
@@ -30,7 +30,7 @@
  "id": "sQbdfyWQhYbO"
  },
  "source": [
- "Uncomment and run the following line in order to install all the necessary dependencies:"
+ "Uncomment and run the following line in order to install all the necessary dependencies (this notebook is being tested with `transformers==4.19.1`):"
  ]
  },
  {
@@ -747,6 +747,7 @@
  " name,\n",
  " evaluation_strategy=\"epoch\",\n",
  " save_strategy=\"epoch\",\n",
+ " logging_strategy=\"epoch\",\n",
  " learning_rate=2e-5,\n",
  " per_device_train_batch_size=batch_size,\n",
  " per_device_eval_batch_size=batch_size,\n",

diff --git a/python/ray/air/examples/huggingface/huggingface_basic_language_modeling_example.py b/python/ray/air/examples/huggingface/huggingface_basic_language_modeling_example.py
@@ -1,6 +1,8 @@
 # Based on
 # huggingface/notebooks/examples/language_modeling_from_scratch.ipynb
 
+# This example is tested with transformers==4.19.1
+
 import argparse
 import tempfile
 
@@ -88,11 +90,12 @@ def train_function(train_dataset, eval_dataset=None, **config):
  training_args = TrainingArguments(
  training_dir,
  evaluation_strategy="epoch",
+ save_strategy="epoch",
+ logging_strategy="epoch",
  num_train_epochs=num_epochs,
  learning_rate=2e-5,
  weight_decay=0.01,
  disable_tqdm=True,
- save_strategy="epoch",
  # Required to avoid an exception
  no_cuda=not torch.cuda.is_available(),
  )

diff --git a/python/ray/train/huggingface/huggingface_trainer.py b/python/ray/train/huggingface/huggingface_trainer.py
@@ -208,6 +208,8 @@ def trainer_init_per_worker(train_dataset, eval_dataset, **config):
  args = transformers.TrainingArguments(
  output_dir=f"{model_checkpoint}-wikitext2",
  evaluation_strategy="epoch",
+ save_strategy="epoch",
+ logging_strategy="epoch",
  learning_rate=2e-5,
  weight_decay=0.01,
  )