Downloads · 30 days
47
6% of all-time downloads
sifat-febo/banglish-companion-3b-mlx
banglish-companion-3b-mlx is a text generation model from sifat-febo. Use it when you need the model to write or continue text. It is set up for mlx. The card lists the license as apache-2.0.
Most models answer romanized Bengali in English, or in Bengali script, or in neither. This one answers the way the message was written.
Downloads · 30 days
47
6% of all-time downloads
All-time downloads
804
Public
Parameters
3.4B
3.7 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors3.6 GB · 100%
How the weights are stored.
U323.4B · 100%
From the Hugging Face model README
Most models answer romanized Bengali in English, or in Bengali script, or in neither. This one answers the way the message was written.
you aj mood bhalo na
model mama mood bhalo na hole patience rakh. Ar ki korle feel better hobe?
3.6 GB, runs on a Mac. Nothing leaves your machine.
Ministral-3-3B, further trained on Banglish conversation. It follows banglish-companion, the 1.7B, and answers shorter and closer to the question.
ki eta? Banglish e kotha bola ekta model, Mac er jonno. 3.6 GB. Internet lagbe na, account lagbe na, taka lagbe na — tumi ja likhba tomar computer er baire jabe na.
ki lagbe? pip install mlx-lm. Apple silicon Mac lagbe.
kemon uttor dey? choto uttor, ar age jigges kore ki hoyeche. lomba lecture dey na. chhoto model — bhul tottho confident bhabe bole, tai fact er jonno eta use koro na.
pip install mlx-lm huggingface_hub
hf download sifat-febo/banglish-companion-3b-mlx --local-dir banglish-3b
python banglish-3b/companion.py
That starts a chat in your terminal. It first answers aj mood bhalo na, the
line shown at the top of this page, then whatever you type. In your own code,
from companion import reply is the only import you need. It reads the
message first with the small ontor classifier that ships here. If someone
writes that they are in danger, it replies with fixed text and does not ask
the model. Otherwise it answers with the settings this model was tested with.
Apple silicon only. For Intel Macs and Linux, the same model in full precision is at banglish-companion-3b.
The author designed the system, chose the base models, ran all training and measurement on the author's own machine, read the models' conversations, and chose what to publish and what to withhold. Claude Code (Anthropic) was used as a coding and writing tool under that direction.
Apache 2.0. Base model Ministral-3-3B-Base-2512 (Mistral AI, Apache 2.0).
Cite the technical report, which describes how the models were built, trained and measured: https://doi.org/10.5281/zenodo.22371896
@misc{febo2026banglishcompanion,
author = {Febo, Sifat},
title = {Banglish Companion: Small Chat Models for Romanized Bengali
— Technical Report},
year = {2026},
month = sep,
publisher = {Zenodo},
version = {1.0},
doi = {10.5281/zenodo.22371896},
url = {https://doi.org/10.5281/zenodo.22371896},
note = {Repository: sifat-febo/banglish-companion-3b-mlx}
}