AI

How the models actually work, what runs on your own hardware, and where the claims stop matching the benchmarks.

Official software

Artificial intelligence · Open source

Ollama Not Using GPU Windows: Diagnose CPU Fallback

Confirm which processor Ollama uses, then fix Windows GPU fallback by checking compatibility, drivers, video memory, model size, and service logs.

Articles

Oct 4, 20267 min

Run LLM Locally on Windows, macOS, or Linux

Check your memory, install Ollama, choose a sensible quantized model, and verify GPU acceleration before sending your first local prompt.

Oct 3, 20268 min

RAG vs fine-tuning: Pick prompt engineering first

Start with prompt engineering, add RAG for changing knowledge, and fine-tune only when the model must consistently learn a task, style, or specialized language.