chunked/nltk_chunking/_llm_tutorial/chunk_43.txt

model_inputs = tokenizer(
     ["1, 2, 3", "A, B, C, D, E"], padding=True, return_tensors="pt"
 ).to("cuda")
generated_ids = model.generate(**model_inputs)
tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
'1, 2, 33333333333'
With left-padding, it works as expected!