sainived656/soreqen-s1
SoreQen S1 by ZorQelis AI is a 2.3 billion parameter bilingual (English/Hinglish) assistant model, fine-tuned from Qwen/Qwen3.5-2B. It specializes in conversational AI, demonstrating improved identity without a system prompt and enhanced Hinglish informative responses. This model is optimized for engaging in natural English and Romanized Hinglish conversations, maintaining the base model's extensive 262,144 token context window and multimodal capabilities.
Loading preview...
SoreQen S1: Bilingual Conversational AI
SoreQen S1, developed by ZorQelis AI, is a 2.3 billion parameter language model fine-tuned from Qwen/Qwen3.5-2B. It is specifically designed for bilingual conversations in English and Romanized Hinglish.
Key Capabilities & Features
- Bilingual Proficiency: Excels in both English and Hinglish (Roman script) conversations, adapting to the user's language and register.
- Enhanced Identity: Demonstrates a strong assistant identity even without an explicit system prompt, indicating its identity is embedded in the model weights.
- Improved Hinglish Informative Responses: Shows better performance in providing informative answers in Hinglish compared to its base model.
- Large Context Window: Inherits the base model's impressive 262,144 token context length, allowing for extended and coherent dialogues.
- Multimodal Capabilities: Retains the vision encoder and multimodal projector from the base Qwen model, though these were frozen during fine-tuning.
- Thinking Mode & Tool Calling: Supports step-by-step reasoning and tool-calling functionalities, inherited directly from the Qwen base model.
What Makes SoreQen S1 Different?
This model's primary differentiator is its specialized fine-tuning for Hinglish conversational fluency and identity. Unlike many general-purpose LLMs, SoreQen S1 focuses on delivering natural and context-aware responses in a specific bilingual context. The fine-tuning process involved a single Low-Rank Adaptation (LoRA) on 21,062 supervised examples, ensuring targeted improvements without altering the base model's core capabilities like vision or structured output.
When to Use SoreQen S1
- Conversational Agents: Ideal for chatbots or virtual assistants requiring fluent interaction in English and Romanized Hinglish.
- Bilingual Support: Applications needing to serve users who communicate in a mix of English and Hindi (Hinglish).
- Identity-Driven Interactions: Scenarios where the AI assistant needs to maintain a consistent and clear persona without constant prompting.
Limitations
- Not suitable for safety-critical or professional advice due to its conversational training.
- Hinglish output is exclusively in Roman script; it does not produce Devanagari.
- As a smaller model, it may confidently state unverified numbers or facts.