And gpt-oss is an offline version of chatgpt

Indeed https://huggingface.co/openai-community

First thing that came to mind: GPT4All

I've been toying with Qwen3.

On my steam deck.

8 bil param model runs stably.

Its's opensource too!

Alpaca is a neat little flatpak that containerizes everything and makes running local models so easy that I can literally do it without a mouse or keyboard.

https://ollama.org/

I mean, most people have a local LLM in their pocket right now.

CrAcKeD

make sure to disconnect the internet first

Le pfffft.

My bet was on porn.

Or a copy of an old Encarta cd-rom

DVDs exist. They can fit approx. 7B params, enough to be somewhat productive.

Qwen3-0.6B is about 400 MB at Q4 and is surprisingly coherent for what it is.

there's also tinyllama, which is somewhere around 600MB. it's hilariously inept. it's like someone jpeg-compressed a robot.

also you're only gonna load off of that cd once so it'll perform fine.

Encarta 95

It could fit on a BDXL disc.

No

(English) 24,05GB without media. Adding media adds 428,36TB.

No, you really can't; It's like 43 gb the text only version

kiwix? that's compressed (afaik), and when i tried, it took up half of my disk space and needed ethernet

The full 2025-04 English-only ZIM dump is about 120 GB. That includes reduced-size images as well as all articles. I think the text-only version is in the 40-60 GB range.

There are smaller ZIM versions in the ~4 GB range that would fit on a DVD, but they're only a subset for specific topics or for a list of the most popular topics.

It's not the LLM that does the web searching, but the software stack around it. On its own, an LLM is just a text completer. What you'd need a frontend like OpenWebUI or Perplexica that would ask the LLM for, say five internet search queries that could return useful information for the prompt, throw those queries into SearxNG, and then pipe the results into the LLM's context for it to be used.

As for the models themselves, any decently-sized one that was released fairly recently would work. If you're looking specifically for open-source rather than open-weight models (meaning that the training data and methodologies were also released rather than just the model weights), GPT-OSS 20B/120B and the OLMo models are recent standouts there. If not, the Qwen3 series are pretty good. (There are other good models out there, this is just what I remember off the top of my head.)

Depends. Does ChatGPT ignore robots.txt too?

I thought it was more like 10-20GB to start out with a usable (but somewhat stupid) model.

Are you confusing the size of the dataset with the size of the model?