align HF tokenization with genai chat scenario in sample test

.github/workflows/causal_lm_cpp.yml

@@ -665,7 +665,7 @@ jobs:
               output.write('question:\n')
               chat_history.append(gen_prompt(prompt))
               chat_prompt = tokenizer.apply_chat_template(chat_history, tokenize=False, add_generation_prompt=True)
-              tokenized = tokenizer(chat_prompt, return_tensors='pt')
+              tokenized = tokenizer(chat_prompt, return_tensors='pt', add_special_tokens=True)
               answer = model.generate(**tokenized, max_length=1000, do_sample=False)
               answer_str = tokenizer.decode(answer[0, tokenized['input_ids'].numel():], skip_special_tokens=True)
               chat_history.append(gen_answer(answer_str))