yanolja/EEVE-Rosetta-4B-FP8-2507

This model is a fine-tuned version of google/gemma-3-4b-pt. As it is intended solely for text generation, we have extracted and utilized only the Gemma3ForCausalLM component from the original architecture.

While the model name includes "EEVE," our well-known model brand, this specific model does not feature an expanded tokenizer. The EEVE branding reflects our commitment to developing high-quality, multilingual models.

Model Name: yanolja/EEVE-Rosetta-4B-FP8-2507
Base Model: google/gemma-3-4b-pt

Model Description

This model is a 4-billion parameter, decoder-only language model built on the Gemma3 architecture and fine-tuned by Yanolja NEXT. It is specifically designed to translate structured data (JSON format) while preserving the original data structure.

The model was trained on a multilingual dataset covering the following languages:

English
Spanish
French
German
Portuguese
Japanese
Korean
Chinese
Arabic
Russian
Hindi

While optimized for these languages, it may also perform effectively on other languages supported by the base Gemma3 model.

How to use

You can use this model with the transformers library as follows:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "yanolja/EEVE-Rosetta-4B-FP8-2507"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype=torch.bfloat16
)

# Example prompt
target_language = "Spanish"
messages = [
    {"role": "system", "content": f"Translate the user's text to {target_language}.\nThink through the translation step by step: first, consider the overall context, then cultural nuances, terminology, initial translation, and self-review.\nAfter this thought process, provide the final translation immediately."},
    {"role": "user", "content": "Yanolja NEXT is a company that provides global cutting-edge technology for the hospitality industry."}
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=4096)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

The model first outputs its thought process within <think> tags, followed by the final {JSON translation}. The output format is as follows:

<think>
though process will be here
</think>

{JSON translation}

Training Procedure

Training Data

The translation datasets were compiled from several sources, including:

To enhance the model's performance with chain-of-thought capabilities, we generated a synthetic reasoning dataset. The process involved:

Using DeepSeek-R1 to translate text from a source to a target language.
Capturing the internal reasoning steps from DeepSeek-R1 only when its translation perfectly matched the ground-truth target text.
Using this collected reasoning data to fine-tune google/gemma-3-27b-it. This fine-tuned model was then used to generate a comprehensive reasoning dataset for training EEVE-Rosetta-4B-2507.

Intended Uses & Limitations

This model is intended for translating structured data (JSON format) while preserving the original structure. It is particularly well-suited for tasks such as localizing product catalogs, translating hotel reviews, or handling any other structured content that requires accurate translation.

Limitations

The model's primary focus is on JSON data. Performance on unstructured text or other data formats may vary.

License

This model is released under the Gemma license, inherited from its base model, google/gemma-3-4b-pt. Please consult the official Gemma license terms for detailed usage guidelines.

Citation

If you use this model, please consider citing:

@misc{yanolja2025eeverosetta,
  author = {Yanolja NEXT},
  title = {EEVE-Rosetta-4B-2507},
  year = {2025},
  publisher = {Hugging Face},
  journal = {Hugging Face repository},
  howpublished = {\\url{https://huggingface.co/yanolja/EEVE-Rosetta-4B-2507}}
}

References

This work utilizes several models and datasets. We would like to acknowledge the original authors for their valuable contributions to the field.

@misc{gemma3,
  author = {Google},
  title = {Gemma 3},
  year = {2024},
  publisher = {Google DeepMind},
  howpublished = {\\url{https://deepmind.google/models/gemma/gemma-3/}}
}

@misc{deepseekai2025deepseekr1incentivizingreasoningcapability,
  title={DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning}, 
  author={DeepSeek-AI},
  year={2025},
  eprint={2501.12948},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2501.12948}, 
}

@misc{aihub,
  author = {National Information Society Agency (NIA)},
  title = {AI-Hub: AI Integrated Platform},
  year = {2025},
  publisher = {National Information Society Agency},
  howpublished = {\\url{https://aihub.or.kr}}
}

@article{europarl,
  author    = {Koehn, Philipp},
  title     = {Europarl: A Parallel Corpus for Statistical Machine Translation},
  journal   = {MT Summit},
  year      = {2005},
  pages     = {79--86}
}