Blogs - Demietrich

Blog Posts

Jul 27, 2026

Do Gemma4-31B or Qwen3.5-122B beat Qwen3.6-27B in any meaningful way?

llm llama.cpp benchmark qwen gemma

Jul 20, 2026

Hands on with the classic number recognition neural net.

MNIST PyTorch deep-learning neural-network fundamentals image-classification

Jul 13, 2026

Fine-tuning small LLMs on consumer hardware: from unsloth CLI to LLaMA-Factory with ZeRO-2.

llm finetuning qwen gemma llama-factory unsloth mediawiki dataset

Jul 12, 2026

An unassuming puzzle game that benchmarks reasoning efficiency well.

llm llama.cpp vllm benchmark qwen gemma

Jul 10, 2026

Four quantizations of Qwen3.6-27B, five capability axes, one dual-3090 rig. The quality question, settled with data instead of vibes.

llm benchmark quantization qwen

Jul 7, 2026

From a simple Ollama container to a dual-RTX 3090 workstation. A weekend experiment to a local LLM obsession.

llm ollama getting-started