Skip to content

Llama 3.3 70B Instruct

A large open-weights text model, and the base for most of the uncensored community fine-tunes people actually use for roleplay.

Open weights
Text
Filtered adult content

At a glance

Made by: Meta
Family: Llama
Licence: Llama 3.3 Community License
Size: 70B parameters
Context: 128000 tokens
Graphics memory: 24 GB minimum, 48 GB recommended
Adult content: FilteredA filter blocks it, with varying success.
On Hugging Face: meta-llama/Llama-3.3-70B-Instruct

The instruct release refuses a great deal. Its value here is as a base: the fine-tunes built on it are what the roleplay community runs, and they are why a 24 GB card is the entry ticket for a quantised copy.

The licence is a community licence with an acceptable-use policy attached, not Apache or MIT. It permits commercial use below a very large user threshold, but it is a licence with terms in it.

Tools that run it

Ollama

One command to download and run a language model on your own machine. The easiest honest answer to "how do I do this without a cloud".
- Self-hosted
- Allowed
- No GPU needed
Run it yourself No GPU needed
Jan

An open-source desktop assistant that runs models locally by default and can be pointed at a hosted one when you choose.
- Hybrid
- Allowed
- No GPU needed
Run it yourself No GPU needed
KoboldCpp

A single-file local model server built for storytelling and roleplay, with the sampler controls that community actually uses.
- Self-hosted
- Allowed
- No GPU needed
Run it yourself No GPU needed
LM Studio

A desktop app for finding, downloading and chatting with local language models. Free, closed source, and the gentlest way in.
- Self-hosted
- Allowed
- No GPU needed
Run it yourself No GPU needed
OpenRouter

One API key for a great many hosted models, including ones with looser content policies than their originals.
- Hosted
- Unknown
Finding models Not known
Text generation web UI

The "AUTOMATIC1111 of language models": one interface, many backends, an extension for everything.
- Self-hosted
- Allowed
- No GPU needed
Run it yourself No GPU needed
llama.cpp

The C++ engine most local language model software is built on. Runs large models on ordinary hardware, including with no GPU at all.
- Self-hosted
- Allowed
- No GPU needed
Run it yourself No GPU needed
vLLM

A serious inference server for serving a model to many people at once. Not a chatbot — the thing a chatbot runs on.
- Self-hosted
- Allowed
Run it yourself 16 GB VRAM, 24 GB recommended