ikhou/dict-xs
The ikhou/dict-xs is a 0.8 billion parameter multilingual dictionary model based on Qwen3-0.6B, fine-tuned by Ikhou on 1.7 million dictionary-style glosses across over 50 languages. It specializes in providing short, context-aware dictionary glosses and translations with grammatical markers for words and phrases. This model is optimized for fast inference in applications requiring quick word lookups, vocabulary learning, and language assistance.
Loading preview...
Ikhou Dictionary Model (dict-xs)
The ikhou/dict-xs is a lightweight, 0.8 billion parameter multilingual dictionary model built upon the Qwen3-0.6B architecture. It has been specifically fine-tuned by Ikhou using 1.7 million dictionary-style glosses covering more than 50 languages. The model's primary function is to generate concise, context-aware translations and grammatical glosses for words and phrases.
Key Capabilities
- Multilingual Support: Provides dictionary glosses for over 50 languages, including major European, Asian, Middle Eastern, and African languages.
- Dictionary-style Output: Generates short, precise glosses with linguistic markers (e.g.,
nm.,nf.,n.,adj.,adv., verb tense information). - Context-Aware: Designed to provide relevant translations based on the surrounding sentence context.
- Fast Inference: Its compact size (596M parameters in bfloat16) allows for rapid processing, making it suitable for real-time applications.
Training Details
The model underwent full fine-tuning on the Qwen3-0.6B base model, utilizing a dataset of 1.7 million synthetic dictionary entries derived from FineWeb and FineWeb-2. The training involved 6,568 steps over approximately 6 hours on NVIDIA H100 hardware, achieving a final loss of 1.30.
Good For
- Reading Applications: Quick word lookups for users reading in foreign languages.
- Vocabulary Learning Tools: Assisting learners with definitions and grammatical information.
- Language Learning Applications: Providing on-demand translation assistance.
- Translation Assistance: Offering short, precise glosses for individual words or short phrases within a given context.
Limitations
While effective for its intended purpose, the model works best with clear, simple contexts and may struggle with proper nouns, very rare languages, or highly ambiguous phrases. It is not intended for sensitive applications or as an authoritative translation tool due to potential biases from its web-trained data.