text-generation-webui: Ubuntu setup and GGUF tutorial

This tutorial walks through text-generation-webui on Ubuntu for private chat with a local LLM. The installation starts with system updates, cloning the repository and running its start script. The speaker selects Nvidia hardware and points out the CPU option, then opens the interface at 127.0.0.1:7860. Password protection is also discussed.

The model download steps differ by format. A standard model needs its Hugging Face repository name; a GGUF download also needs the exact quantized filename. The speaker uses SmolLM3 3B first and describes it as suitable for modest hardware. For GGUF loading, the walkthrough covers the llama.cpp loader and GPU layer controls that divide work between CPU and GPU.

The settings section covers temperature, saved parameter presets and custom system instructions. Character cards add a reusable persona and example conversations. Chat controls include instruction modes, token counts and conversation branching. The speaker reports errors when loading vision models and leaves image chat untested in the tutorial.

After finding the first model less suitable for conversation in these tests, the speaker switches to an Unsloth GGUF of gpt-oss-20b. This example uses an 8192-token context and adjusts GPU layers before loading. It demonstrates email generation, the interface's displayed reasoning text and a follow-up request to summarize the email. GPU layers and context length must be set before loading in the demonstrated workflow.