phogpt-0.13b

from transformers import AutoModelForSeq2SeqLM
from huggingface_hub import hf_hub_download
import sentencepiece as spm
import torch

model = AutoModelForSeq2SeqLM.from_pretrained("amaury-delille/phogpt-0.13b", trust_remote_code=True)

en_spm_path = hf_hub_download("amaury-delille/phogpt-0.13b", "tokenizer_en/spm.model")
vi_spm_path = hf_hub_download("amaury-delille/phogpt-0.13b", "tokenizer_vi/spm.model")

en_tokenizer = spm.SentencePieceProcessor(en_spm_path)
vi_tokenizer = spm.SentencePieceProcessor(vi_spm_path)

text = "Hello, how are you?"
encoded = en_tokenizer.Encode(text)
encoded.append(3)
max_len = 128
encoded = encoded + [0] * (max_len - len(encoded))
input_ids = torch.tensor([encoded[:max_len]], dtype=torch.long)

model.eval()
outputs = model.generate(input_ids, max_length=128)

out_tokens = [t for t in outputs[0].tolist() if t not in [0, 2, 3]]
translation = vi_tokenizer.Decode(out_tokens)
print(f"English: {text}")
print(f"Vietnamese: {translation}")