BeeLLM

BeeLLM

Offers a family of tiny GGUF language models that talk in bee buzzes, from buzz-only bees to mood-tagged ones and linked queen and worker hives. For hobbyists and developers who like small models.

Made by schizophyllume on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: BeeLLM is a family of tiny language models that answer in bee noises. The smallest models only buzz, the Mood models add feelings, and the hive models connect a queen bee with her workers. Every file is a llama-architecture GGUF, so it runs wherever llama.cpp runs and is small enough to sit inside a web app.

Who it suits: It suits people who enjoy small language-model oddities, and developers who want a tiny model for testing. Teachers can use it to show how a language model works at a small scale.

Industry | Who uses it | How they use it | Example
Education | Teachers and students | Show how a language model behaves at tiny scale | Comparing a one-layer tiny bee with the eight-layer Scintilla
Web development | Front-end developers | Run a small model inside a web application | Loading a GGUF bee file in a web page
Entertainment | Hobby makers | Give a chat character moods and small actions | Building a pet bee that reacts to how you treat her
Research | Small-model tinkerers | Study a queen model linked to worker models | Trying the files in the hive folder

If you need a model that gives factual answers, you can skip it, since these bees reply in buzzes, moods and small actions.

Where to use it:

• llama.cpp on Windows, macOS or Linux
• A web application, since the files are small enough to run there
• A browser page the author hosts for chatting with the bees

Use case | Who | What they get back
Chat with a moody bee | Anyone curious | A reply that opens with a mood tag, then buzzing and actions
Test a tiny model | Developers | A GGUF file with as few as 10K weights
Try linked models | Researchers | A hive file holding a queen and up to 12 workers

How to set it up:

• Python, for the Hugging Face download tool
• A runtime for GGUF files, such as llama.cpp

1. Install the Hugging Face hub package.
2. Download the model files.
3. Pick a bee from the tiny, mood or hive folder.
4. Load it in llama.cpp or in your own web application.

pip install -U huggingface_hub
hf download schizophyllume/BeeLLM

How to use it: Load one bee file in your GGUF runtime and send it a message. Tiny bees answer with buzzing only. Mood bees open every reply with a tag shaped like mood and level. The mood is neutral, happy, loving, sad, angry or confused, and the level is low or high. The tag follows how you treat her, so a flower makes her happy and a swat makes her angry.

Tool | What it does
tiny folder | Buzz-only bees with a vocabulary of 8 to 21 buzzes
mood folder | Bees that tag each reply with a mood and answer with feelings and actions
BeeLLM-Scintilla-50M | The 50M hive queen on her own, whose mood changes gradually
hive folder | A 50M queen linked to up to 12 worker bees, plus a small first hive

Example: You send "i love you" to a Mood bee. By the card's reaction table, she feels loving, so her reply opens with a loving tag and carries on in buzzes and small actions.

For newcomers and experts: A language model guesses what comes next in a text. These ones are tiny, and the smallest bee has a vocabulary of just 8 buzzes. A good first try is to load the smallest tiny file and say hello, or to chat with a Mood bee on the author's browser page.

Under the hood, every file is a llama-architecture model with a 4096 context. The tiny bees have one layer and 10K to 24K weights. The Mood bees have 4 to 8 layers and 1.1M to 50M weights. In the hive files, the queen and her workers pass neural activations to each other, which the card calls neuralese. Small size has a cost, and the card says the 2.0 1M bee garbles a word now and then.

You send a message
The bee model reads it
A mood tag is picked from how you act
The reply comes in buzzes and actions

Good to know:

Title: Vocabulary size per bee
BeeLLM-30P: 8 buzzes
BeeLLM-42P: 12 buzzes
BeeLLM-42P-Chaos: 21 buzzes
BeeLLM-1M-Mood: 1,024 tokens
BeeLLM-2.0-12M-Mood: 2,048 tokens
BeeLLM-Scintilla-50M: 8,192 tokens

• Some tiny bees can only buzz, and they keep buzzing when you buzz at them.
• The Scintilla bee remembers her mood, so after you are mean she recovers one step at a time.
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: none named, since the models run from a downloaded file
• Not for: answering real questions, since the bees reply in buzzes, moods and small actions

On Hugging Face, 13 November 2026

• Task: text generation
• Parameters: 4.4M
• Weights: about 2 MB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 1,850
• Likes: 46
• License: apache-2.0
• Quantized versions: 2
• Tags: gguf, llama.cpp, tiny, toy-model, swarm, text-generation

Links

• Model: https://huggingface.co/schizophyllume/BeeLLM
• Files: https://huggingface.co/schizophyllume/BeeLLM/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download schizophyllume/BeeLLM

Then follow the usage in its model card. Nothing in this kit runs by itself.