eduardem/parrot_en_es_13B_v2
The eduardem/parrot_en_es_13B_v2 is a 13 billion parameter LLaMa-2-based model fine-tuned by eduardem for English to Spanish in-context translation. This second iteration focuses on context-aware translations, preserving original formatting, and is suitable for various applications like mobile apps. It was trained with 250,000 real examples, utilizing input masking and a 32-bit Adam optimizer for improved accuracy and efficiency.
Loading preview...
Overview
The eduardem/parrot_en_es_13B_v2 is the second iteration of an English to Spanish in-context translation model, fine-tuned from a LLaMa-2-13B base. Developed by eduardem, this model is specifically designed to provide accurate and relevant translations by understanding and utilizing the context provided in the user prompt. It was trained on a dataset of 250,000 real examples.
Key Enhancements in V2
This version introduces several significant improvements over its predecessor:
- New Codebase: Fine-tuned using an entirely new codebase for enhanced performance and reliability.
- Input Masking: Implements input masking during training to exclude input from the loss ratio, leading to more precise translations.
- 32-bit Adam Optimizer: Upgraded from an 8-bit to a 32-bit Adam optimizer, improving learning efficiency and convergence speed.
- Increased Training: Utilizes a bigger batch size and more epochs during fine-tuning.
Core Features
The model offers distinct capabilities for translation tasks:
- Context-Aware Translations: Delivers translations tailored to specific contexts or themes.
- Formatting Preservation: Maintains original formatting, including line breaks, HTML, and XML, ensuring integrity of the translated text.
- Versatile Application: Suitable for a wide range of contexts, from mobile applications to other general uses.
Usage and Prompt Template
To leverage the model effectively, users must adhere to a specific prompt template. This template guides the model to translate English text into Spanish within a defined context, while strictly preserving formatting. The model's output provides the desired translation in the first block, with any subsequent text considered extraneous.
Ideal Use Cases
This model is particularly well-suited for scenarios requiring:
- Precise, Contextual Translation: When the meaning of a word or phrase changes significantly based on its surrounding context (e.g., "tap" in a mobile app vs. a dance guide).
- Technical or Structured Text Translation: For content where preserving formatting like HTML, XML, or line breaks is crucial.
- Application Localization: Translating UI strings or content for applications where context and formatting are paramount.