add a limit to the max_input_length to get rid of the OOM issues.

dalm/datasets/qa_gen/question_answer_generation.py

@@ -64,12 +64,18 @@ def parse_args() -> argparse.Namespace:
         action="store_true",
         help="Save the files as CSV. If False, will save them as a dataset directory via [`~Dataset.save_to_disk`]",
     )
+    parser.add_argument(
+        "--max_input_tokens",
+        type=int,
+        default=512,
+        help="Maximum number of input tokens for the model.",
+    )
     args = parser.parse_args()
     return args
 
 
 def generate_question_answer_pairs(
-    documents: dict, model: PreTrainedModel, tokenizer: PreTrainedTokenizer, passage_column_name: str
+    documents: dict, model: PreTrainedModel, tokenizer: PreTrainedTokenizer, passage_column_name: str, max_input_tokens: int
 ) -> dict:
 
     """Generate question answer pairs"""
@@ -120,7 +126,7 @@ def generate_question_answer_pairs(
         add_generation_prompt=True
     )
 
-    model_inputs = tokenizer(batch_texts, return_tensors="pt", padding=True, truncation=True).to(model.device)
+    model_inputs = tokenizer(batch_texts, return_tensors="pt", padding=True, truncation=True, max_length=max_input_tokens).to(model.device)
 
     generated_ids = model.generate(
         **model_inputs,
@@ -179,7 +185,7 @@ def split_dataset(
 
 
 def generate_qa_from_dataset(
-    dataset: Dataset, passage_column_name: str, title_column_name: str, sample_size: int, batch_size: int, load_in_8bit: bool = True
+    dataset: Dataset, passage_column_name: str, title_column_name: str, sample_size: int, batch_size: int, max_input_tokens: int, load_in_8bit: bool = True
 ) -> DatasetDict:
     logger.info(f"Generating question answer pairs with batch size: {batch_size}")
     tokenizer = AutoTokenizer.from_pretrained(QA_MODEL)
@@ -197,7 +203,7 @@ def generate_qa_from_dataset(
         f"Test dataset size: {len(small_dataset_splits['test'])}"
     )
     qa_gen_map = partial(
-        generate_question_answer_pairs, model=model, tokenizer=tokenizer, passage_column_name=passage_column_name
+        generate_question_answer_pairs, model=model, tokenizer=tokenizer, passage_column_name=passage_column_name, max_input_tokens=max_input_tokens
     )
     processed_data = small_dataset_splits.map(qa_gen_map, batched=True, batch_size=batch_size)
     # Print all questions from the test split before filtering
@@ -254,9 +260,10 @@ def generate_qa_from_disk(
     batch_size: int,
     output_dir: str,
     as_csv: bool,
+    max_input_tokens: int,
 ) -> None:
     dataset = _load_dataset_from_path(dataset_path)
-    qa_gen_data = generate_qa_from_dataset(dataset, passage_column_name, title_column_name, sample_size, batch_size)
+    qa_gen_data = generate_qa_from_dataset(dataset, passage_column_name, title_column_name, sample_size, batch_size, max_input_tokens)
     output_path = Path(output_dir)
     output_path.mkdir(exist_ok=True)
     for split_name, split_ds in qa_gen_data.items():
@@ -279,6 +286,7 @@ def main() -> None:
         args.batch_size,
         args.output_dir,
         args.as_csv,
+        args.max_input_tokens,
     )
 
 
@@ -290,5 +298,6 @@ def main() -> None:
     --dataset_path=knowledge_dataset.csv \
     --batch_size=8 \
     --sample_size=50 \
-    --output_dir=out
+    --output_dir=out \
+    --max_input_tokens=512
 """