This model is a specialized, style-constrained version of KvaytG/marian-mt-en-ru-high-precision, fine-tuned for a very specific use case: aggressive, informal dialogue translation in forced uppercase.

The primary focus of this version is emotional impact, conversational "thou" (ты) addressing, and strict adherence to a capitalized output format.

Model Description

The model was fine-tuned on a curated subset of 1,000,000 sentence pairs extracted from conversational corpora (OpenSubtitles, Tatoeba).

Key Features of Data Preparation:

Length Constraint: Only short, punchy phrases (strictly ≤ 10 words) were used to maintain high dialogue density.
Style Filtering: The dataset was filtered to prioritize informal speech patterns, replacing formal Russian "Вы" with the familiar "Ты".
Case Normalization: Both source and target sides of the training corpus were converted to UPPERCASE to ensure the model learns to operate within this specific character space.

Final Result

The model successfully transitioned from the surgical, academic precision of its predecessor to a bold, informal tone. It consistently outputs translations in uppercase and correctly maps English neutral pronouns to Russian informal equivalents.

Usage

Note: For best results, it is recommended to pass the input string in .upper() as the model was trained on uppercase tokens.

python

import torch
from transformers import MarianMTModel, MarianTokenizer

model_name = "KvaytG/marian-mt-en-ru-informal-caps"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)


def translate(text):
    input_text = text.upper()
    inputs = tokenizer(input_text, return_tensors="pt", padding=True)
    with torch.no_grad():
        output_tokens = model.generate(**inputs, max_new_tokens=64)
    return tokenizer.decode(output_tokens[0], skip_special_tokens=True)


print(translate("Shut up and look at me!"))
# Expected: ЗАТКНИСЬ И ПОСМОТРИ НА МЕНЯ!

print(translate("How are you doing today?"))
# Expected: КАК У ТЕБЯ ДЕЛА СЕГОДНЯ?

License

This model is released under the Apache License 2.0.

Citation

bibtex

@misc{kvaytg_marian_mt_en_ru_informal_caps,
  author       = {KvaytG},
  title        = {MarianMT English-Russian model for informal dialogue in forced uppercase},
  year         = {2026},
  publisher    = {Hugging Face},
  journal      = {Hugging Face Models},
  url          = {https://huggingface.co/KvaytG/marian-mt-en-ru-informal-caps},
  note         = {Fine-tuned from KvaytG/marian-mt-en-ru-high-precision (itself based on Helsinki-NLP/opus-mt-en-ru) for informal/colloquial style with ALL CAPS output.}
}