metacognitive-behavioral-tuning/Qwen3-1.7B-gpt-oss-distill
The metacognitive-behavioral-tuning/Qwen3-1.7B-gpt-oss-distill model is a 2 billion parameter language model based on the Qwen3-1.7B architecture, fine-tuned using a gpt-oss-distill method. This model specializes in multi-hop question answering, leveraging a distillation process from a larger teacher model (gpt-oss-120b) followed by GRPO. It is specifically optimized for complex question answering tasks, demonstrating capabilities on benchmarks like HotpotQA, MuSiQue, and 2WikiMultiHopQA. With a context length of 32768 tokens, it is designed for processing and reasoning over extensive information to answer intricate queries.
Loading preview...
Overview
The metacognitive-behavioral-tuning/Qwen3-1.7B-gpt-oss-distill is a 2 billion parameter language model built upon the Qwen/Qwen3-1.7B base architecture. Its development involved a unique gpt-oss-distill method, which includes naive distillation of raw traces from a teacher model (gpt-oss-120b) via Supervised Fine-Tuning (SFT), subsequently enhanced with GRPO (Generative Reinforcement Learning with Policy Optimization).
Key Capabilities
- Multi-Hop Question Answering: Specifically designed and optimized for complex question answering tasks that require reasoning across multiple pieces of information.
- Distillation-based Learning: Leverages knowledge distilled from a more powerful teacher model, enabling strong performance within a smaller parameter count.
- Benchmarked Performance: Evaluated on multi-hop question answering datasets such as HotpotQA (in-domain) and MuSiQue / 2WikiMultiHopQA (out-of-domain), indicating its proficiency in this domain.
Good For
- Research in Multi-Hop QA: Ideal for researchers exploring advanced techniques in multi-hop question answering and knowledge distillation.
- Applications Requiring Complex Reasoning: Suitable for use cases where models need to synthesize information from various sources to answer intricate questions, particularly within a 32768 token context window.
- Efficient Deployment: As a 2 billion parameter model, it offers a balance between performance in specialized tasks and computational efficiency compared to much larger models.