from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model = AutoModelForCausalLM.from_pretrained(
    "SORNPov/Chamnaot2.5",
    dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("SORNPov/Chamnaot")

question = "សរសេរចំណោទនៅទីនេះ (your Khmer math question here)"
prompt = f"question: {question} guidance: "

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

output = model.generate(
    **inputs,
    max_new_tokens=800,
    do_sample=False,
    pad_token_id=tokenizer.eos_token_id
)

input_len = inputs["input_ids"].shape[1]
print(tokenizer.decode(output[0][input_len:], skip_special_tokens=True))