import re
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
    "laion/anh-xglm-7.5b-cross-lingual",
)
tokenizer = AutoTokenizer.from_pretrained(
    "laion/anh-xglm-7.5b-cross-lingual",
)
whitespace_tokens_map = {'\n': '<n>', '  ': '<w>'}
text = "User: Apa yang terjadi pada pertempuran Cannae? Jawab dalam bahasa China.\n"
for k, v in whitespace_tokens_map.items():
    text = text.replace(k, v)
inputs = tokenizer(text, return_tensors="pt")
tokens = model.generate(**inputs)
output = tokenizer.decode(tokens[0], skip_special_tokens=True)
for v in whitespace_tokens_map.values():
    output = re.sub(rf"{v}\s+(\S+)", rf"{v}\1", output)
for k, v in whitespace_tokens_map.items():
    output = output.replace(v, k)