Below is the Python code to run the model. It automatically utilizes the GPU if available and loads the model from the local cache after the first run.

python

import torch
from transformers import AutoTokenizer, AutoModel

# 1. Configuration
repo_id = "myduy/dlm-vi2en-checkpoint-61500"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 2. Load Model & Tokenizer
# trust_remote_code=True is required for custom architectures
tokenizer = AutoTokenizer.from_pretrained(repo_id, trust_remote_code=True)
model = AutoModel.from_pretrained(repo_id, trust_remote_code=True).to(device)
model.eval()

# 3. Prepare Input
text = "cậu có muốn đến nghe không?"
inputs = tokenizer(text, return_tensors="pt").to(device)

# 4. Generate
with torch.no_grad():
    output_tokens = model.generate(
        inputs.input_ids, 
        max_iterations=50, 
        temperature=1.0, 
        strategy="reparam-uncond-deterministic-cosine"
    )

# 5. Decode Output
output_text = tokenizer.batch_decode(output_tokens, skip_special_tokens=True)[0]

print(f"Input: {text}")
print(f"Output: {output_text}")