Your question is Making a Small LLM Reason. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do you make a SLM to be good at reasoning?
Describe and implement an end-to-end approach for improving reasoning in a small language model. Address training-data construction, curriculum design, supervised fine-tuning, preference or verifier-based optimization, evaluation against reasoning benchmarks, contamination, hallucination, and inference-time latency. Explain how you would distinguish memorization from genuine multi-step reasoning and how you would monitor regressions after deployment.