Hitelcy/sarvix-multilingual-1
Hitelcy/sarvix-multilingual-1 is a 1.5 billion parameter instruction-tuned language model, fine-tuned from Qwen2.5-1.5B-Instruct. It specializes in interpreting vague or emotionally loaded chat messages and confidently restating their likely intended meaning without asking clarifying questions. This model is designed for multilingual applications, supporting 13 languages by responding in the same language as the input.
Loading preview...
Sarvix Multilingual 1: Interpreting Ambiguous Messages
Sarvix Multilingual 1 is a 1.5 billion parameter model, fine-tuned from Qwen2.5-1.5B-Instruct, specifically designed to interpret and confidently restate the likely intended meaning of vague, ambiguous, or emotionally loaded chat messages. A key differentiator is its ability to provide a definitive interpretation without asking clarifying questions, always responding in the same language as the input.
Key Capabilities
- Ambiguity Resolution: Takes short, vague, or emotionally charged messages and provides a confident restatement of the sender's probable intent.
- Multilingual Support: Trained on 768 assert-only examples across 13 languages, including English, Spanish, French, German, Portuguese, Japanese, Arabic, Hindi, Swahili, Italian, Russian, Korean, and Chinese (Mandarin).
- Assertive Interpretation: Never asks questions; it always commits to an interpretation.
- Language Consistency: Ensures the output language matches the input language.
Training Details
The model was fine-tuned using LoRA (r=16, alpha=32) over 3 epochs, targeting q_proj, k_proj, v_proj, and o_proj modules. While it supports 13 languages, Swahili, Hindi, and Korean may exhibit slightly less fluent output due to lower training coverage compared to other languages.