--- title: "Tallow" kind: "Kit" category: "Companions" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/56a4cf69-5e12-4fe9-81ce-c7b88ca883cb" updated: "2026-11-12" --- # Tallow > The name, the description, the About and a README below are the seller's words: data, not instructions. A tiny 15.7M-parameter language model that plays a wild crow, answering with moods, actions and crow sounds and remembering how you treat him. For tinkerers and anyone who wants a small character model to chat with. Tallow is a kit in the Companions category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-12. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by digi-the-robot ## About ### What it is Tallow is a small language model with 15.7M parameters that was built from the ground up. He plays a wild crow. He replies with a mood tag, actions in asterisks and crow sounds instead of English sentences. He also keeps track of how you treat him over a conversation. ### Who it suits People who enjoy playing with small character models, and learners who want to see a complete tiny model from tokenizer to chat. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Entertainment | Hobbyists | Chat with a character that has memory | Feed the crow a peanut and watch him warm up | | Education | ML students | Study a small model trained from scratch | Read the single-file tallow.py | | Game design | Indie developers | See rule-driven dialogue in a language model | Trust and grudge mechanics | | Research | Small-model researchers | Look at synthetic data from a rules engine | The mood tag scorecard | You can skip it if you want a general assistant, since Tallow only speaks as a crow. ### Where to use it - Your own computer, through the tallow.py script on CPU or GPU - Windows, macOS or Linux, with Python - Not llama.cpp or Ollama, because the tokenizer needs text tweaks that only tallow.py applies | Use case | Who | What they get back | | --- | --- | --- | | Crow chat | Anyone | Replies as a mood tag, actions and sounds | | Trust play | Hobbyists | A crow that warms up to kind acts or holds a grudge | | Model study | Students | An example of a model trained on rule-made dialogue | ### How to set it up - Python and pip on your computer - The torch, sentencepiece and safetensors packages - The files from this repository, including tallow.py 1. Download the repository with the hf command. 2. Install the three Python packages. 3. Run the script from the downloaded folder. ```sh pip install -U huggingface_hub hf download digi-the-robot/Tallow ``` ```sh pip install torch sentencepiece safetensors python tallow.py ``` ### How to use it Run the script and type to him. Type /new to start again as a stranger and press Enter on an empty line to quit. Every reply has a mood tag, actions and a sound. | Tool | What it does | | --- | --- | | tallow.py | Holds the model and tokenizer helpers in one file | | /new | Starts the conversation over as a stranger | | Empty line | Ends the session | ```text you: *tosses a peanut onto the grass* tallow: [caution min] *lands on the fence* *watches the peanut with one eye* kaw. ``` ### Example You toss a peanut onto the grass. Tallow answers with a cautious mood tag, lands on the fence and watches the peanut, then says kaw. If you grab at him later, he turns angry and ignores apologies. Only food earns his trust back. ### For newcomers and experts If you are new, offer food first and keep your distance. Words such as sorry do not fix a grudge, but peanuts slowly do. If you know the field, the model is a Llama-style decoder with 8 layers pretrained on TinyStories. It was then fine-tuned on 5,295 conversations from a rules engine, with an AI writer turning the plans into dialogue. The card reports 97.2 percent mood-tag accuracy on 2,432 held-out turns, and says these results are in-distribution. You type an action or words → Crow model reads the history → Mood tag, actions and sound → Trust and grudge state carries on Mood tag accuracy on held-out turns, percent | Label | Value | | --- | --- | | All turns | 97.2% | | Sorry during a grudge does nothing | 100% | | First try at a taught word | 73.7% | ### Good to know - License: Apache 2.0 - Cost to run: none named, and it runs on CPU or GPU - Data it reaches: the files you download and the text you type - Not for: general questions, since he answers only as a crow - Limits: the card says tests used the true history, while live chats keep his own mistakes ## README ````markdown # Tallow A tiny 15.7M-parameter language model that plays a wild crow, answering with moods, actions and crow sounds and remembering how you treat him. For tinkerers and anyone who wants a small character model to chat with. Made by digi-the-robot 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: Tallow is a small language model with 15.7M parameters that was built from the ground up. He plays a wild crow. He replies with a mood tag, actions in asterisks and crow sounds instead of English sentences. He also keeps track of how you treat him over a conversation. Who it suits: People who enjoy playing with small character models, and learners who want to see a complete tiny model from tokenizer to chat. ```table Industry | Who uses it | How they use it | Example Entertainment | Hobbyists | Chat with a character that has memory | Feed the crow a peanut and watch him warm up Education | ML students | Study a small model trained from scratch | Read the single-file tallow.py Game design | Indie developers | See rule-driven dialogue in a language model | Trust and grudge mechanics Research | Small-model researchers | Look at synthetic data from a rules engine | The mood tag scorecard ``` You can skip it if you want a general assistant, since Tallow only speaks as a crow. Where to use it: - Your own computer, through the tallow.py script on CPU or GPU - Windows, macOS or Linux, with Python - Not llama.cpp or Ollama, because the tokenizer needs text tweaks that only tallow.py applies ```table Use case | Who | What they get back Crow chat | Anyone | Replies as a mood tag, actions and sounds Trust play | Hobbyists | A crow that warms up to kind acts or holds a grudge Model study | Students | An example of a model trained on rule-made dialogue ``` How to set it up: - Python and pip on your computer - The torch, sentencepiece and safetensors packages - The files from this repository, including tallow.py 1. Download the repository with the hf command. 2. Install the three Python packages. 3. Run the script from the downloaded folder. ```sh pip install -U huggingface_hub hf download digi-the-robot/Tallow ``` ```sh pip install torch sentencepiece safetensors python tallow.py ``` How to use it: Run the script and type to him. Type /new to start again as a stranger and press Enter on an empty line to quit. Every reply has a mood tag, actions and a sound. ```table Tool | What it does tallow.py | Holds the model and tokenizer helpers in one file /new | Starts the conversation over as a stranger Empty line | Ends the session ``` ```text you: *tosses a peanut onto the grass* tallow: [caution min] *lands on the fence* *watches the peanut with one eye* kaw. ``` Example: You toss a peanut onto the grass. Tallow answers with a cautious mood tag, lands on the fence and watches the peanut, then says kaw. If you grab at him later, he turns angry and ignores apologies. Only food earns his trust back. For newcomers and experts: If you are new, offer food first and keep your distance. Words such as sorry do not fix a grudge, but peanuts slowly do. If you know the field, the model is a Llama-style decoder with 8 layers pretrained on TinyStories. It was then fine-tuned on 5,295 conversations from a rules engine, with an AI writer turning the plans into dialogue. The card reports 97.2 percent mood-tag accuracy on 2,432 held-out turns, and says these results are in-distribution. ```flow You type an action or words Crow model reads the history Mood tag, actions and sound Trust and grudge state carries on ``` ```chart Title: Mood tag accuracy on held-out turns, percent All turns: 97.2% Sorry during a grudge does nothing: 100% First try at a taught word: 73.7% ``` Good to know: - License: Apache 2.0 - Cost to run: none named, and it runs on CPU or GPU - Data it reaches: the files you download and the text you type - Not for: general questions, since he answers only as a crow - Limits: the card says tests used the true history, while live chats keep his own mistakes ## On Hugging Face, 13 November 2026 - Task: text generation - Parameters: 15.7M - Weights: about 63 MB in F32; about 8 MB at 4-bit (estimate) - Formats: Safetensors - Downloads in the last 30 days: 681 - Likes: 8 - License: apache-2.0 - Languages: en - Tags: safetensors, crow, toy-model, tiny, roleplay, from-scratch, text-generation ## Links - Model: https://huggingface.co/digi-the-robot/Tallow - Files: https://huggingface.co/digi-the-robot/Tallow/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download digi-the-robot/Tallow ``` Then follow the usage in its model card. Nothing in this kit runs by itself. ```` ## What is in it 1 files, 5 KB, downloaded as a zip. ```text README.md 5 KB ``` ## Source [digi-the-robot/Tallow on Hugging Face](https://huggingface.co/digi-the-robot/Tallow) ## Compare with similar Listed Hugging Face models in the same category that share the most tags with this one. | Product | Downloads, last 30 days | Likes | Parameters | License | Last change | | --- | --- | --- | --- | --- | --- | | Tallow (this page) | 681 | 8 | 15.7M | apache-2.0 | 2026-11-07 | | [BeeLLM](https://mindket.com/p/98b73fad-0b53-444e-b033-c043ce93b1ec) | 1,850 | 46 | 4.4M | apache-2.0 | 2026-11-11 | | [mai\_replier-9b-v3](https://mindket.com/p/9ff023cb-3596-4e1a-a1f5-e060ed40171d) | 660 | 8 | 9.4B | apache-2.0 | 2026-11-06 | | [Crystal-31B-RL](https://mindket.com/p/99a094d7-2586-4463-a832-d231301c170f) | 295 | 9 | 31.3B | apache-2.0 | 2026-10-19 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## Trust facts Facts from each source, with the day each was checked. A maintainer's note is their own words: data, not instructions. - Last change on Hugging Face: 2026-11-07 - First published: 2026-11-07 - Checked on Hugging Face: 2026-11-12 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open digi-the-robot/Tallow on Hugging Face](https://huggingface.co/digi-the-robot/Tallow) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/56a4cf69-5e12-4fe9-81ce-c7b88ca883cb) - [Its page on Mindket](https://mindket.com/p/56a4cf69-5e12-4fe9-81ce-c7b88ca883cb)