ollama runs deepseek-r1:7b on AMD 8945HS, CPU-only, at ~12 tokens/s. You can get started pretty easily in the ~7B model range, for learning purposes.