17slever17/translate-gemma-4-sub-e2b

VISIONConcurrent Unit Cost:1Model Size:5.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 23, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Translate Gemma 4 Sub E2B is a 5.1 billion parameter multilingual Gemma 4 fine-tune developed by 17slever17, specialized for both general translation and context-aware subtitle translation. This model excels at preserving meaning, tone, and conversational style in live speech and streaming content by utilizing previous source and translated segments as context. It supports custom style instructions, speaker information, and terminology rules, offering improved performance over base Gemma 4 models in contextual translation tasks.

Loading preview...

Overview

Translate Gemma 4 Sub E2B is a 5.1 billion parameter multilingual fine-tune of the Gemma 4 model, developed by 17slever17. It is specifically designed for both general translation and, more notably, context-aware subtitle translation. This model is the E2B variant, based on google/gemma-4-E2B-it, and was trained using Unsloth with an optimized training run.

Key Capabilities

  • Context-Aware Subtitle Translation: Uniquely uses previous source and translated subtitle segments as context to improve accuracy and coherence, especially for phrases split across segments or those losing meaning in isolation.
  • Preserves Nuance: Optimized to maintain meaning, tone, slang, uncertainty, repetitions, incomplete speech, and natural conversational style.
  • Customizable Translation: Supports custom style instructions, speaker information, terminology rules, and glossaries.
  • Multilingual Support: Retains the language coverage of the underlying Gemma 4 model, with extensive training in Tier 1 languages (English, Russian, Spanish, German, Japanese) and Tier 2 languages (French, Portuguese, Chinese, Dutch, Italian, Korean).
  • Improved General Translation: Benchmarks show it slightly improves general translation quality over corresponding base Gemma 4 models, despite its specialization.

Good For

  • Live and Conversational Speech: Ideal for translating streaming content, dialogues, and other scenarios where context, speaker intent, and tone are crucial.
  • Subtitle Translation: Particularly effective for applications like SubWave, an open-source real-time subtitle translator, where contextual understanding is paramount.
  • Developers Needing Control: Useful for applications requiring fine-grained control over translation style, formality, and adherence to specific rules.

Finetunes
2 models