Tag: ollama

API vs self-hosted LLM cost reality, GPU TCO, privacy, latency, break-even math. When paying OpenAI/Anthropic wins. When local wins.

Local coding agents don't fail because your 27B model is too small. They fail because you let 200K tokens of garbage pile up in the context window. Cap it low.

Self-host Gemma 4 or Qwen3-Coder via Ollama or llama.cpp in Docker, then let Claude delegate the grunt work to it. Free tokens, zero code leakage.

Local LLMs can call tools, query APIs, and run code if you set them up right. Function calling on Ollama and llama.cpp explained, patterns that actually work.

Gemma 4 vs Qwen3.6: sizes, reasoning, coding benchmarks, and which model you should actually pull for your home lab rig.

Pixtral, Qwen3-VL, and Gemma 4 compared for local multimodal use in 2026. LLaVA is dead; here's what to run in Ollama for OCR, screenshots, and vision tasks.

How tiny 7B and 8B models keep punching above their weight, knowledge distillation, the teacher-student trick that makes local AI actually usable on home hardware.

Google's Gemma 4 is the best open model they've shipped yet. Here's how to pull it, run it, and actually use it for real work with Ollama on your own hardware.

Before you download a 70B model, calculate if it fits. The formulas, the gotchas, and a quick calculator you can actually use.

vLLM, llama.cpp, and Ollama all run local LLMs, compare throughput, memory use, GPU support, and which fits your hardware.

Ollama can load one model at a time on limited hardware. How to switch between models, use CPU offloading, and manage VRAM intelligently.

Most people use OpenAI's embeddings because it's easy. But local embeddings exist. How to pick and when it actually matters.