muradil211/AetherSearch_SFT
muradil211/AetherSearch_SFT is a 3.1 billion parameter Qwen2.5-3B-Instruct based causal language model fine-tuned by muradil211. It is specifically designed as a compact search agent that learns to reason, retrieve, and answer by generating search queries and processing retrieved information. This model excels at integrating external search capabilities into its reasoning process, making it suitable for retrieval-augmented generation tasks.
Loading preview...
What is AetherSearch SFT?
AetherSearch SFT is a 3.1 billion parameter model, fine-tuned from Qwen2.5-3B-Instruct, designed to act as a compact search agent. It learns to reason, retrieve information, and formulate answers by interacting with an external search or RAG backend. The model was trained on 2,000 complete search trajectories, including both single-search and multi-search examples, to develop its search-native behavior.
Key Capabilities
- Search-native behavior: The model learns to determine when and what to search for before generating an answer.
- Evidence-in-the-loop reasoning: It integrates retrieved passages as context for reasoning, allowing for evidence-grounded final answers.
- Compact and accessible: Built on a 3B parameter backbone, it's suitable for experimentation and deployment.
- Reproducible: Training code, data checksums, and schema tests are publicly available.
How it Works
The model generates <think>, <search>, and <answer> spans. Your runtime is responsible for executing the <search> query, retrieving information, and returning it to the model within <information> tags. This allows the model to iteratively search and refine its understanding.
Good for
- Developing agentic search applications.
- Implementing retrieval-augmented generation (RAG) systems where the model actively formulates search queries.
- Experimenting with compact models for complex reasoning and retrieval tasks.