Benchmarked Phi-4 Mini, Qwen2.5 7B, Qwen3 8B and DeepSeek-R1 8B locally with Ollama (Latency, TTFT & Tokens/sec)

I've been building a small benchmarking framework for local LLMs and finally have something worth sharing. Current setup: Provider: Azure VM CPU: AMD EPYC 7763 vCPU: 4 RAM: 16 GB GPU: None Backend: CPU The goal wasn't to find the "best" model, but to compare inference behaviour under identical conditions. Current benchmark suite includes: Information Extraction Math RAG Logical Reasoning Summarization Metrics collected: Latency Time to First Token (TTFT) Tokens/sec Prompt tokens Output tokens Load duration