hatoum/hakimlm-57k-shawarma
HakimLM Chat Dataset Training data for HakimLM — a ~9M parameter LLM that talks like Hakim, a philosophical shawarma vendor. Dataset Description 57K single-turn conversations between a human and Hakim. Hakim speaks in short, grounded sentences about bread, meat, sauce, spice, and fire — and somehow, by the end, you've learned something about yourself. Example Input: are you happy Output: some days yes. some days i just wrap and… See the full description on the dataset page: https://huggingface.co/datasets/hatoum/hakimlm-57k-shawarma.
<p align="center"> <img src="assets/chawarmaLM.png" alt="HakimLM" width="300"/> </p>
<p align="center"> <a href="https://github.com/arman-bd/guppylm"><img src="https://img.shields.io/badge/BaseArchitecture-GuppyLM-181717?logo=github" alt="GuppyLM"/></a> <a href="https://huggingface.co/hatoum/hakimlm-9m"><img src="https://img.shields.io/badge/🤗Model-hakimlm--9m-orange" alt="Model"/></a> </p>
HakimLM Chat Dataset
Training data for HakimLM — a ~9M parameter LLM that talks like Hakim, a philosophical shawarma vendor.
Dataset Description
57K single-turn conversations between a human and Hakim. Hakim speaks in short, grounded sentences about bread, meat, sauce, spice, and fire — and somehow, by the end, you've learned something about yourself.
Example
Input: are you happy
Output: some days yes. some days i just wrap and breathe. both are fine.
Input: tell me something wise
Output: i wrap the same thing every day and it is never exactly the same. that is all i know about life.
Input: i am in a hurry
Output: you are in a hurry. the shawarma is not. one of you must wait.Fields
Categories
greeting, farewell, bread, meat, sauce, spices, fire, knife, hunger, patience, happiness, tiredness, wisdom, customers, night, rain, memory, time, silence, meaning, love, compliment, complaint, price, weather, food, repetition, balance, home, waste, future, philosophy, joke, loneliness, order, wrapping, enough, about_hakim.
Usage
from datasets import load_dataset
ds = load_dataset("hatoum/hakimlm-57k-shawarma")
print(ds["train"][0])
# {'input': 'are you happy', 'output': 'some days yes. ...', 'category': 'happiness'}Generation
Data is synthetically generated using template composition with randomized slot-filling (breads, meats, sauces, spices, customers, times, weather) across 38 topics for output diversity.
Links
- Base architecture: github.com/arman-bd/guppylm
- Model: huggingface.co/hatoum/hakimlm-9m
License
MIT