lmt-60-1.7b-en-vi

LMT Translation Model (EN ↔ VI)

This is a merged model combining the NiuTrans LMT-60-1.7B base model with fine-tuned LoRA adapters for English-Vietnamese translation.

Model Details

• Base Model: NiuTrans/LMT-60-1.7B
• Adapter Path: ./model_lmt/checkpoint-20000
• Task: Bidirectional Translation (English ↔ Vietnamese)
• Model Type: Causal Language Model (Qwen3) with merged LoRA weights

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    "Ripefog/lmt-60-1.7b-en-vi",
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("Ripefog/lmt-60-1.7b-en-vi", trust_remote_code=True)

# Translate English to Vietnamese
messages = [
    {"role": "user", "content": 'Dịch câu sau sang tiếng Việt: "Hello, how are you?"'},
]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

with torch.no_grad():
    outputs = model.generate(
        inputs,
        max_new_tokens=256,
        num_beams=5,
        do_sample=False,
    )

response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)
print(response)

Translation Directions

English to Vietnamese (EN → VI)

messages = [
    {"role": "user", "content": 'Dịch câu sau sang tiếng Việt: "Your English text here"'},
]

Vietnamese to English (VI → EN)

messages = [
    {"role": "user", "content": 'Translate the following sentence into English: "Văn bản tiếng Việt của bạn ở đây"'},
]

Training Details

This model was fine-tuned using LoRA (Low-Rank Adaptation) on translation datasets. The LoRA adapters have been merged with the base model for easier deployment and faster inference.

License

Apache 2.0