VelsTech AI
From understanding how LLMs work to running models on your own hardware – a natural learning path, in plain language, without the hype.
Understand the basics
New to this? Follow the guided path →
What is machine learning, really?
What learning from data actually means – training, models, and prediction – before the LLMs.
What is a Large Language Model, really?
A plain-language look at how LLMs work under the hood – tokens, context, and probabilities.
How to write better prompts
Practical patterns that get better answers from ChatGPT, Claude, and similar tools.
Local AI
How to get started with local AI (2026)
Pick hardware, install Ollama, download a model, and run it on your own machine.
GGUF explained
What the file format is, what Q4_K_M means, and how to pick the right quantization.
llama.cpp guide
Install, run, and benchmark GGUF models with the engine behind Ollama.
Local vs cloud AI
Privacy, cost, and capability – running models on your own hardware versus cloud APIs.
How much VRAM do you need?
VRAM for 7B, 14B, and 32B models – weights, quantization, KV cache, and context.
Best GPU for running LLMs locally
RTX vs Radeon, how much VRAM you really need, and what actually matters for llama.cpp.
Tested on real hardware
Benchmark database
Search, filter, and compare tokens/sec for LLMs on real GPUs – tested and estimated.
Qwen 27B: ROCm vs Vulkan on RX 6800M
18 vs 22 tok/s decode at 16K on a 12 GB card – tested in the Lab.
Ornith 35B MoE at 262K: ROCm vs Vulkan
26 vs 20 tok/s decode on a 12 GB card with 28 CPU experts.
MoE vs Dense on RX 6800M
3B active vs 27B at 16K/262K – which architecture wins where.
Crunch your own numbers
Model news & updates
Go under the hood
Run on your hardware
ROCm & Vulkan on AMD
Which backend to pick, how to install, and benchmark results on the RX 6800M.
Running LLMs on Apple Silicon
How Macs handle local AI, what models fit your RAM, and getting started with Metal.
RAG from scratch
Ask questions about your own documents – load, embed, retrieve, generate, all local.