import torch
from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model="etamin/Letz-MT-gemma2-2b-en-lb",
    model_kwargs={"torch_dtype": torch.bfloat16},
    device="cuda",  # replace with "mps" to run on a Mac device
)

messages = [
    {"role": "user", "content": """
          <|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nYou are a helpful AI assistant for translation.<|eot_id|>
          <|start_header_id|>user<|end_header_id|>\n\nTranslate the English input text into Luxembourgish.
          Do not include any additional information or unrelated content.\n\n
          "I have not met you yet."
          """},
]

outputs = pipe(messages, max_new_tokens=256)
assistant_response = outputs[0]["generated_text"][-1]["content"].strip()
print(assistant_response)
# ech sinn em d'éinescht nach begéint

Translation Template

The instruction-tuned models use a chat template that must be adhered to for conversational use. The easiest way to apply it is using the tokenizer's built-in chat template, as shown in the following snippet.

Let's load the model and apply the chat template to a conversation. In this example, we'll start with a single user interaction:

from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch

model_id = "etamin/Letz-MT-gemma2-2b-en-lb"
dtype = torch.bfloat16

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="cuda",
    torch_dtype=dtype,)

chat = [
    { "role": "user", "content": """
          <|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nYou are a helpful AI assistant for translation.<|eot_id|>
          <|start_header_id|>user<|end_header_id|>\n\nTranslate the English input text into Luxembourgish.
          Do not include any additional information or unrelated content.\n\n
          "I have not met you yet."
          """ },
]
prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)

After the prompt is ready, generation can be performed like this:

inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
outputs = model.generate(input_ids=inputs.to(model.device), max_new_tokens=150)
print(tokenizer.decode(outputs[0]))

Inputs and outputs

Input: Text string, such as a prompt to do the translation task.
Output: Generated Luxembourgish text in response to the input.

Citation

none

@misc{song2025llmsilverbulletlowresource,
      title={Is LLM the Silver Bullet to Low-Resource Languages Machine Translation?}, 
      author={Yewei Song and Lujun Li and Cedric Lothritz and Saad Ezzini and Lama Sleem and Niccolo Gentile and Radu State and Tegawendé F. Bissyandé and Jacques Klein},
      year={2025},
      eprint={2503.24102},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.24102}, 
}

Model Data

Data used for model training and how the data was processed.

Training Dataset

RTL (https://www.rtl.lu/) Lod.lu (https://data.public.lu/en/datasets/letzebuerger-online-dictionnaire-lod-linguistesch-daten/)

mirix/etamin-Letz-MT-gemma2-2b-en-lb-GGUF

Gemma 2 English to Luxembourgish model card

Model Information

Description

Usage

Running with the pipeline API

Translation Template

Inputs and outputs

Citation

Model Data

Training Dataset

mirix/
etamin-Letz-MT-gemma2-2b-en-lb-GGUF

Running with the `pipeline` API