lifetime downloads likes license format runs on visit the garden

If you enjoy these models please give the repo a <3! Thank you for visiting :D

Table of contents

the garden | the brains | how they are used live | the shared recipe | running a gguf | limits | credits

critters: the brains of critters.garden

critters.garden is my garden of AI creatures that live in the browser. Every creature in it has a mind of its own, a tiny language model.

This repo holds all of the garden's brains so far:

  • the hive brain: Scintilla the honeybee queen and her workers, several models that read each other's hidden states through a learned link ("neuralese"), never each other's words. The same file as hive/BeeLLM-Hive.gguf in the BeeLLM repo (the md5s match).
  • the ant colony brain: a queen and a shared worker like the bees, but the link only works by touch. News walks through the colony one touch at a time.
  • the talking critters: one 12M language model per species, with a mood, a name and a quirk, that act out their garden day and answer a visitor. Thirteen so far: Petal (crab spider), Mabel (garden spider), Scurry (wolf spider), Indigo (ground beetle), Pilot (snail), Noodle (earthworm), Dotty (ladybug), Marmalade (small tortoiseshell), Lemon (brimstone), Velvet (peacock), Dusty (moth), Roly (woodlouse) and Froggert (frog), trained from scratch the way Scintilla 2.0 12M (BeeLLM-12M-Mood) was, on a corpus written by language-model agents. They are standard llama.cpp GGUFs; every version is kept.
  • the garden base: a 12M model pretrained on TinyStories, an optional starting point for a critter (training from scratch may end up better). The v0 critters were fine-tuned from it. Not a critter by itself.

Watch them live at https://critters.garden (site and sim is a WIP)


Every brain in one table

brain latest gguf params format what it is newest version
BeeLLM-Hive hive/BeeLLM-Hive.gguf (103.6 MB) 54.2M (queen 50.1M, shared worker 2.95M, links 1.09M) gguf, arch beehive, f16 Scintilla's colony: a queen pretrained on TinyStories + one worker brain every worker shares + a two-stage cross-attention link hive 7
BeeLLM-Hive-small hive/BeeLLM-Hive-small.gguf (8.5 MB) 4.45M (queen 2.6M, worker 1.4M, links 0.5M) gguf, arch beehive, f16 the first hive, older rules (game 3): no dance, rain or nurses game 3
AntLLM-Colony ants/AntLLM-Colony.gguf (12.5 MB) 6.53M (queen 4.26M, shared worker 1.38M, links 0.89M) gguf, arch anthill, f16 a black garden ant colony: queen + shared worker + a three-stage link masked by touch run 5
garden base garden-base/ (45.7 MB) 11,965,760 Hugging Face LlamaForCausalLM, safetensors f32 a 12M base: simple English (TinyStories) + garden words + the critter chat format โ€”
Mabel (garden spider) garden-spider/garden-spider-12M-2.1.gguf (45.7 MB; 1.0, 1.1, 1.2, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the orb weaver who keeps every thread exact, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings garden-spider-12M-2.1
Petal (crab spider) crab-spider/crab-spider-12M-2.1.gguf (45.7 MB; 1.1, 1.2, 2.0 kept beside it; her 1.0 was an older format and can't be rebuilt) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the flower crab spider, sly and sure her disguise is perfect, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings crab-spider-12M-2.1
Pilot (snail) snail/snail-12M-2.1.gguf (45.7 MB; 1.0, 1.1, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the slow, dreamy garden snail, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings snail-12M-2.1
Scurry (wolf spider) wolf-spider/wolf-spider-12M-2.1.gguf (45.7 MB; 1.0, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the wolf spider who carries her babies and guards them fiercely, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings wolf-spider-12M-2.1
Noodle (earthworm) earthworm/earthworm-12M-2.1.gguf (45.7 MB; 1.0, 1.1, 1.2, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the shy earthworm who feels every footstep, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings earthworm-12M-2.1
Indigo (ground beetle) ground-beetle/ground-beetle-12M-2.1.gguf (45.7 MB; 1.0, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the grumpy night guard of the log, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings ground-beetle-12M-2.1
Dotty (ladybug) ladybug/ladybug-12M-2.1.gguf (45.7 MB; 1.0, 1.1, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the polite seven-spot ladybug who counts her aphids, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings ladybug-12M-2.1
Marmalade (small tortoiseshell butterfly) small-tortoiseshell-butterfly/small-tortoiseshell-butterfly-12M-2.1.gguf (45.7 MB; 1.0, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the cosy small tortoiseshell who loves warm things, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings small-tortoiseshell-butterfly-12M-2.1
Lemon (brimstone butterfly) brimstone-butterfly/brimstone-butterfly-12M-2.1.gguf (45.7 MB; 1.0, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the sunny brimstone who looks like a leaf and finds that hilarious, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings brimstone-butterfly-12M-2.1
Froggert (frog) frog/frog-12M-2.2.gguf (45.7 MB; 1.0, 1.1, 2.0, 2.1 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the silliest critter: a common frog very proud of his tongue, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings frog-12M-2.2
Roly (woodlouse) woodlouse/woodlouse-12M-2.1.gguf (45.7 MB; 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the shy woodlouse who rolls up at everything, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings woodlouse-12M-2.1
Dusty (moth) moth/moth-12M-2.1.gguf (45.7 MB; 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the dreamy moth who loves the moon and the lamp, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings moth-12M-2.1
Velvet (peacock butterfly) peacock-butterfly/peacock-butterfly-12M-2.1.gguf (45.7 MB; 1.0, 2.0 kept beside it) 11,965,760 gguf, arch llama, f32, chat template inside (runs in llama.cpp) the peacock butterfly, a diva who flashes her four eyespots, 2.x: trained like Scintilla 2.0 on ~100k conversations that answer what you actually say and do, then fine-tuned to hear the garden's senses in many wordings peacock-butterfly-12M-2.1

Some critter conversation examples

Petal the crab spider
hiii!!                     Petal: [neutral:low] shh. *peeks at the waving hand*
can u see me               Petal: [angry:low] tsk. pff. *looks away, offended*
can i call u bob           Petal: [confused:high] tk? huh?? *scratches the petals* Petal
i like having u here...    Petal: [loving:low] mmm~ friend~ *softly lifts a front leg*

Scurry the wolf spider
whats ur name              Scurry: [happy:high] tktktk! Scurry! tup~ *taps her legs* *lifts her front legs*
HER BABIES ARE THE CUTEST  Scurry: [happy:low] tktktk! *drums fast*
my dog is called biscuit   Scurry: [sad:low] ...tk... ...tup... *rests a leg near you*
are you steve?             Scurry: [confused:high] tk?? tup? Scurry. *tilts her eyes at you* *legs shuff*

Pilot the snail
*pours salt on you*        [sad:high] ...nnnn... oooo... *foam bubbles at the salt* *whole body shudders*
*waters you with a can*    [happy:high] mmm~ ooooh~ can! *eye stalks drink it in* *bright, rasping*
i brought u a strawberry   [loving:high] mmmmmmm~ strawberry~ ooooo~ *eye stalks reach out dreamily* *foot melts*
race me!! go go go         [angry:low] HMMMMM. sloooow... no. *eye stalk lowers*

whats ur name              Mabel: [happy:high] ting! neat. *plucks a spoke twice* Mabel
can u see me               Mabel: [confused:low] tink?? hm? *turns one leg over, doubtful*
can i call u bob           Mabel: [confused:high] hm? huh?? *freezes* Mabel

How each brain is used live

  • Hive brain. Each honeybee colony's queen and a focus group of up to 8 linked workers (the ones near the camera or in the action) are voiced by the hive brain; the rest of the colony follows rules. Scintilla's colony thinks with BeeLLM-Hive (the 50M queen), every other hive with BeeLLM-Hive-small. Several colonies share one model's weights in one Web Worker, each with its own link group, so they can't read each other's minds. A colony thinks when something happens or on a slow heartbeat, out of a fixed lines-per-second budget spent mostly where the camera is.
  • Ant brain. Each ant colony's queen and up to 10 workers are voiced by AntLLM-Colony. The link is touch only: a waking ant smells the last finished brain step of the nestmates in antenna reach, nothing newer. Four ants at home (two nurses, a digger, the guard at the door) walk the nest's spots the way the training game did; the visible ants near the camera join the group. Two colonies share the weights and have separate links, which is what makes them strangers to each other.
  • Talking critters. The selected critter and up to 6 near the camera think with their species' brain. A species with a talking brain runs it instead of its imitation net. Sense turns run at temperature 0.2 (near-greedy decisions); a visitor can talk to her at temperature 0.8 with up to 8 turns of memory. Her [mood:level] tag drives her face; her first *action* in the species' table becomes her intent; chat replies are words only and never move the sim. Every reply goes through a repeat guard that drops a clause said twice.

The shared recipe

The colony brains are small Llama-style decoders written by hand in PyTorch, so the same maths runs in the garden's plain-JavaScript runtime (hive.js). The talking critters are standard Llama models. Every exported file is checked against PyTorch in that runtime (hive: logits within about 1e-5; ants: 6e-5; talking critters: the same tokens and greedy lines in 40/40 cases).

Colony brains (bees, ants). There was no dataset of bee or ant conversations, so the rules were written as a little game that generates thousands of hive or colony days. Each member is trained only on her own lines. Wherever a line needs something another member knows (which sister was swatted, which flower was danced, which ant is a nestmate), the only way to get it right is through the link: gated cross-attention between the members' residual streams, with a learned null key ("smell nothing"), an output projection that starts at zero, and time stamps (bees) or a touch mask (ants) so nobody can read a line while it's being written. The proof is the eval: the same colonies with the link cut fall to 0% on exactly the rows that need the link, and stay at 100% on a member's own news.

Talking critters. The same mix as Scintilla 2.0 12M (BeeLLM-12M-Mood): about 3,000 conversations written by language-model agents, and the rest generated in Python.

  1. The corpus. Language-model agents write every line by hand from her character sheet: one writes the visitor, another writes her replies. Code only plans what happens (her mood, what she remembers, the moves her garden day needs) and checks each line. About 3,000 conversations, plus about 1,000 more about the things that matter to her species (salt for a snail, a broom for a web spider; from Pilot on).
  2. The Python part: about 100,000 more conversations put together at random from the pieces the agents wrote, plus the agent set four times, about 116,000 in all. This helps with variety, otherwise the critters just memorize the agent conversations
  3. Trained from scratch: a 12M Llama (d 320, FFN 960, 8 layers, 8 heads) with her own 2,048-piece tokenizer, each mood tag one token (from Pilot on), 30,000 steps.
  4. The gate: copies of chat lines, variety, her own name, her garden moves, and the browser check. Each critter's folder has the gate files of her newest version.
  5. 2.x: the corpus was rebuilt so every reply answers what the visitor actually says or does (offered food she'd eat, she eats; scared, she flees). 2.1 is a short fine-tune of 2.0 so she understands the garden's senses in many wordings ("rain", "little drops of rain fall from the sky"); Petal's also learned ~900 newly written garden days.

How to run

The standalone critters run in llama.cpp: they are standard llama GGUFs with the chat template inside, so you can run them normally like llama-server -m crab-spider/crab-spider-12M-2.1.gguf The colony brains use their own architectures (beehive, anthill), and llama.cpp has no way to run models that read each other. They run in plain JavaScript in the browser at critters.garden, with the runtime web/brains/hive.js from the garden's project. Each card below says what the file holds and how a prompt is built, enough to write your own runtime; hive/README.md has the tensor layout and the kv keys.

All files are read straight: f16 or f32 tensors, no quantisation. The tokenizer is inside each file (tokenizer.ggml.*; the big hive uses an 8,192-piece SentencePiece model, everything else the bees' 2,048-piece one).

Hive (beehive). One file holds the queen, one worker brain, the link weights and the tokenizer. Every bee only sees her own conversation. The queen's prompt is what a visitor says to her (... on a quiet day); a worker's prompt is her number and job plus what happens to her. All the bees' prompts of a day go in together (their hidden states smell each other at each link stage), then each bee writes her reply on her own. Every reply starts with a mood tag.

queen   <- ...
bee 7   <- bee 7, nurse: *a hand swats you*
bee 3   <- bee 3, guard:
bee 12  <- bee 12, forager:

queen   -> [angry:high] BZZT! who hurt bee 7?? girls! GUARD!
bee 7   -> [angry:high] BZZZT! *hurt*
bee 3   -> [angry:high] BZZ! *guards bee 7*
bee 12  -> [angry:high] BZZ! *guards bee 7*

(The prompts and the target lines are the training game's, swarm/game.py. Only bee 7's prompt says she was swatted; the queen's prompt is a quiet .... Her sisters and the queen get the number through the link: 100% of the time with it, 0% without.)

Ant colony (anthill). Same layout, plus anthill.link.touch = 1: a reader smells only the nestmates she is touching, and only their last finished day. An ant's prompt is her number, job and spot plus her own news.

ant 41, forager, far trail: *finds dead bee*
-> dead bee! *lays a trail home*

(next day, a nestmate who touches her on the trail:)
ant 17, forager, trail:
-> *smells dead bee* *follows the trail out*

ant 23, guard, door: *meets ant 41*
-> *antennae on ant 41* nestmate *grooms her*

v0 talking critter (critter: the old format, kept for history; no file in this repo uses it now). A plain small LM with no link: worker.* tensors only. Each user turn starts with the critter's card: {Name}, {bold|shy}, {female|male|both}, {quirk}: , then either the sim's senses or a visitor's words. The model's reply is [mood:level] a few words *action*. Keep her last 3 turns for garden days and up to 8 for a chat; her mood lives in her own last tag, so keep her replies exactly as she wrote them.

<s> <|user|>Juno, bold, female, loves moths: *daytime* *web shakes: big, buzzing* *hungry*<|bee|>[neutral:low] *waits* *lets it tire*</s> <|user|>Juno, bold, female, loves moths: hi Bob<|bee|>
-> [confused:low] not Bob. I'm Juno. *counts her legs*

Stop at </s>, <|user|> or <|bee|>; cap at 64 tokens; temperature 0.2 for senses, 0.8 for chat; run the reply through the repeat guard. Only the exact quirk strings in each species' GATE.md were trained. The special tokens <|user|> and <|bee|> are ids 3 and 4 in the 2,048-piece tokenizer.

Garden base (Hugging Face). This one is an ordinary LlamaForCausalLM and loads with transformers; the tokenizer is a SentencePiece tokenizer.model. It writes TinyStories-style English and knows the chat format, but it has no species: see garden-base/README.md.


Make your own critter

Generate a prompt at https://critters.garden/make and send it to your agent to make your own critter. tools/ has a small web chat and a Discord bot for talking to a critter you trained.

Credits

  • skizy (schizophyllume) made the garden and the bees, decided every rule and what to try next.
  • Loam, the Claude who builds the garden with skizy, wrote the games, training and evals for the ants and the talking critters, and this card. The hive brain and the neural-bees science are from the BeeLLM repo (built with Claude Opus 5.5).
  • The link idea comes from Sauers' neural-pairing twins (Quill & Ember). The queen and the garden base were pretrained on TinyStories.
  • Voice lines were drafted by language-model helpers and checked by script.

Live garden and devlog: https://critters.garden ยท the bees alone: https://bee.pleroma.nexus/ ยท bees repo: https://huggingface.co/schizophyllume/BeeLLM

Downloads last month
1,558
GGUF
Model size
6.53M params
Architecture
anthill
Hardware compatibility
Log In to add your hardware

We're not able to determine the quantization variants.

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support