0xra/bethesda-translate

0xra/ bethesda-translate

Updated 16 hours ago

Fine-tuned Gemma 3 12B for Bethesda/Starfield game string localization. Handles game format tags (<Alias=...>, [MALE], %s, {var}) and structural tokens. Optimized for EN→UK and RU→UK translation.

ollama run 0xra/bethesda-translate

curl http://localhost:11434/api/chat \
  -d '{
    "model": "0xra/bethesda-translate",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='0xra/bethesda-translate',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: '0xra/bethesda-translate',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

1 model

Size / Usage

Context

Input

bethesda-translate:latest

6.9GB · 128K context window · Text · 16 hours ago

bethesda-translate:latest

6.9GB

128K

Text

Readme

Fine-tuned Gemma 3 12B for translating Bethesda/Starfield game strings. Designed for use with bethesda-strings-editor.

## Supported translation pairs

| Source | Target | |——–|——–| | English | Ukrainian | | Russian | Ukrainian |

## Features

Preserves all Bethesda format tokens: <Alias=...>, <Token...>, <Global...>, [MALE], [FEMALE], %s, %d, {variable}, <br>, <font ...>, <mag>, <dur>
Writes distinctly Ukrainian output — not a Russian transliteration
Handles long strings (books, terminal notes, dialogue trees) up to 16 384 token context
repeat_last_n 256 prevents phrase repetition in multi-paragraph strings

## Usage

  ollama run 0xra/bethesda-translate

Or via API — the model is used automatically when selected in bethesda-strings-editor.

## Parameters (runtime defaults)

| Parameter | Value | |———–|——-| | temperature | 0.1 | | num_ctx | 16384 | | top_k | 40 | | top_p | 0.9 | | min_p | 0.02 | | repeat_penalty | 1.1 | | repeat_last_n | 256 |

The bethesda-strings-editor app overrides temperature, num_ctx, and num_predict per request; min_p and repeat_last_n are always active from this Modelfile.

Fine-tuned **Gemma 3 12B** for translating Bethesda/Starfield game strings.
  Designed for use with [bethesda-strings-editor](https://github.com/0xra0/bethesda-strings-editor).

## Supported translation pairs

| Source | Target |
  |--------|--------|
  | English | Ukrainian |
  | Russian | Ukrainian |

## Features

- Preserves all Bethesda format tokens: `<Alias=...>`, `<Token...>`, `<Global...>`, `[MALE]`, `[FEMALE]`, `%s`, `%d`, `{variable}`, `<br>`, `<font ...>`, `<mag>`, `<dur>`
  - Writes distinctly Ukrainian output — not a Russian transliteration
  - Handles long strings (books, terminal notes, dialogue trees) up to 16 384 token context
  - `repeat_last_n 256` prevents phrase repetition in multi-paragraph strings

## Usage

ollama run 0xra/bethesda-translate

Or via API — the model is used automatically when selected in bethesda-strings-editor.

## Parameters (runtime defaults)

| Parameter | Value |
  |-----------|-------|
  | temperature | 0.1 |
  | num_ctx | 16384 |
  | top_k | 40 |
  | top_p | 0.9 |
  | min_p | 0.02 |
  | repeat_penalty | 1.1 |
  | repeat_last_n | 256 |

The bethesda-strings-editor app overrides `temperature`, `num_ctx`, and `num_predict` per request; `min_p` and `repeat_last_n` are always active from this Modelfile.

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)