Minor fixes of finetuning examples

kasnerz · Jun 1, 2023 · 33755f8 · 33755f8
1 parent fc6b0a4
commit 33755f8
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 30 deletions.
diff --git a/examples/README.md b/examples/README.md
@@ -2,6 +2,10 @@
 
 You can load the datasets in a unified way for training the models.
 
+
+## Requirements
+Please, first install the additional requirements listed in `./requirements.txt`.
+
 ## Fine-tuning using `transformers` library
 `finetuning_transformers.py` is an example of fine-tuning a seq2seq model and running inference using `transformers` library. It can be run from the command line, e.g. :
 ```
@@ -23,12 +27,3 @@ The example `multitasking.py` is almost equivalent to `finetuning_transformers.p
 Dataset-specific task description is prepended to each input item before training. <br>
 In this example, custom linearization functions are implemented for E2E and WebNLG datasets.
 
-
-## Requirements
-
-Requirements to run the examples (in addition to `tabgenie`):
-* `numpy`
-* `evaluate`
-* `transformers==4.25.1`
-* `torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113`
-* `sacrebleu`
diff --git a/examples/finetuning_transformers.py b/examples/finetuning_transformers.py
@@ -16,14 +16,6 @@
 from tabgenie import load_dataset
 
 
-# extra requirements:
-# numpy
-# evaluate
-# transformers==4.25.1
-# torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
-# sacrebleu
-
-
 SEED = 42
 random.seed(SEED)
 np.random.seed(SEED)
@@ -36,7 +28,7 @@
 
 MAX_LENGTH = 512
 LABEL_PAD_TOKEN_ID = -100
-PATIENCE = 5
+PATIENCE = 3
 
 BLEU_METRIC = evaluate.load("sacrebleu")
 
@@ -77,8 +69,8 @@ def compute_bleu(eval_preds, tokenizer):
 @click.command()
 @click.option("--dataset", "-d", required=True, type=str, help="Dataset to train on")
 @click.option("--base-model", "-m", default="t5-small", type=str, help="Base model to finetune")
-@click.option("--epochs", "-e", default=30, type=int, help="Maximum number of epochs")
-@click.option("--batch-size", "-b", default=16, type=int, help="Path to the output directory")
+@click.option("--epochs", "-e", default=10, type=int, help="Maximum number of epochs")
+@click.option("--batch-size", "-b", default=16, type=int, help="Number of examples in a batch")
 @click.option("--ckpt-dir", "-c", default=os.path.join(ROOT_DIR, "checkpoints"), type=str, help="Directory to store checkpoints")
 @click.option("--output-dir", "-o", default=os.path.join(ROOT_DIR, "models"), type=str, help="Directory to store models and their outputs")
 def main(dataset, base_model, epochs, batch_size, ckpt_dir, output_dir):

diff --git a/examples/multitasking.py b/examples/multitasking.py
@@ -17,14 +17,6 @@
 from tabgenie import load_dataset
 
 
-# extra requirements:
-# numpy
-# evaluate
-# transformers==4.25.1
-# torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
-# sacrebleu
-
-
 SEED = 42
 random.seed(SEED)
 np.random.seed(SEED)
@@ -37,7 +29,7 @@
 
 MAX_LENGTH = 512
 LABEL_PAD_TOKEN_ID = -100
-PATIENCE = 5
+PATIENCE = 3
 
 BLEU_METRIC = evaluate.load("sacrebleu")
 
@@ -92,7 +84,7 @@ def compute_bleu(eval_preds, tokenizer):
 @click.command()
 @click.option("--datasets", "-d", required=True, type=str, help="Datasets to train on")
 @click.option("--base-model", "-m", default="t5-small", type=str, help="Base model to finetune")
-@click.option("--epochs", "-e", default=30, type=int, help="Maximum number of epochs")
+@click.option("--epochs", "-e", default=10, type=int, help="Maximum number of epochs")
 @click.option("--batch-size", "-b", default=16, type=int, help="Path to the output directory")
 @click.option("--ckpt-dir", "-c", default=os.path.join(ROOT_DIR, "checkpoints"), type=str, help="Directory to store checkpoints")
 @click.option("--output-dir", "-o", default=os.path.join(ROOT_DIR, "models"), type=str, help="Directory to store models and their outputs")