decode_input_ids.py

import sys
import torch
from transformers import AutoTokenizer

input_ids=[32, 720, 103, 432, 146, 167, 1400, 195, 294, 1083, 111]
input_ids=[1,  88,  64, 779,   2,  64,   1,  89,  64]
model = "/sgl-workspace/guppylm-9M"
tokenizer = AutoTokenizer.from_pretrained(
    model,
    trust_remote_code=True,
)
text = tokenizer.decode(
    torch.tensor(input_ids, dtype=torch.long),
    skip_special_tokens=False,
)
sys.stdout.write(text)