Supervised fine-tuning (SFT) trajectories across four search-intensive benchmarks: SimpleQA, FRAMES, WebWalkerQA, and Seal0. These trajectories facilitate the training of agentic systems for long-horizon information retrieval and synthesis, specifically targeting Small Language Models (SLMs).
Use Cases
- Fine-tune Small Language Models (SLMs) for long-horizon information retrieval using the SFT trajectories
- Improve model accuracy on the SimpleQA benchmark by training on the provided search-intensive data
- Develop agentic synthesis capabilities for complex research tasks using the synthesis-focused training examples
Strengths
- Includes SFT trajectories for the Fathom-DeepResearch agentic system
- Targets performance on the FRAMES, WebWalkerQA, SimpleQA, and Seal0 benchmarks
- Designed for the development of the Fathom-Search-4B model