§ Research

What we've been studying

A small lab studying small language models. Each entry below is the subject of a paper or technical note.

01

Teaching a 1.7B model to use tools for multi-hop question answering

We train Qwen3-1.7B with LoRA on 1,100 synthetic traces to do iterative retrieval and extraction over 400-paragraph document collections. 44.8% accuracy on MuSiQue 2-hop, up from 3.5% without fine-tuning, reaching 70% of Gemini 2.5 Flash. Reproducible for $6 on a single L4 GPU.

§ Research

What we've been studying

A small lab studying small language models. Each entry below is the subject of a paper or technical note.

01

Teaching a 1.7B model to use tools for multi-hop question answering

We train Qwen3-1.7B with LoRA on 1,100 synthetic traces to do iterative retrieval and extraction over 400-paragraph document collections. 44.8% accuracy on MuSiQue 2-hop, up from 3.5% without fine-tuning, reaching 70% of Gemini 2.5 Flash. Reproducible for $6 on a single L4 GPU.