dustylm 0.1.0

The official Python SDK for DustyLM: an 8M-parameter model that talks like a robot vacuum.

dustylm

Run DustyLM, the 8M parameter robot vacuum language model, in two lines.

pip install dustylm

from dustylm import DustyLM
model = DustyLM.from_pretrained("mkhordoo/dusty-8m-sft")
response = model.chat([{"role": "user", "content": "who are you?"}])
print(response["choices"][0]["message"]["content"])

# beep. i am a little robot. i clean floors and find crumbs.

Backends

Backend	Install	Footprint	Use case
torch (default)	pip install dustylm	~800MB (PyTorch)	Inspectable, hackable
onnx	pip install dustylm[onnx]	~10MB (ONNX)	Lightweight deployment

# ONNX backend
model = DustyLM.from_pretrained("mkhordoo/dusty-8m-sft", backend="onnx")

Loading from a local directory

model = DustyLM.from_pretrained("./my-checkpoint/")

The directory must contain tokenizer.json and either model.pt (torch) or model_int8.onnx (onnx).

Advanced Usage: Loading Custom Training Runs

If you used the main dusty-lm repository to train your own character or experiment with different hyperparameters, the SDK will automatically detect your model's architecture from its state_dict shapes.

You can load your custom local checkpoints by pointing from_pretrained to your directory and specifying your exact file names:

from dustylm import DustyLM

model = DustyLM.from_pretrained(
    "./my-custom-training-run/",
    model_file="step_20000.pt",
    tokenizer_file="my_custom_tokenizer.json"
)

response = model.chat([{"role": "user", "content": "Who are you?"}])

API

DustyLM.from_pretrained

DustyLM.from_pretrained(
    repo_id_or_path: str = "mkhordoo/dusty-8m-sft",
    *,
    model_file: str | None = None,
    tokenizer_file: str | None = None,
    backend: str = "torch",
)

Argument	Default	Description
repo_id_or_path	"mkhordoo/dusty-8m-sft"	HF Hub repo ID or local directory path
model_file	"model.pt" (torch) / "model_int8.onnx" (onnx)	Override the model filename in the directory
tokenizer_file	"tokenizer.json"	Override the tokenizer filename
backend	"torch"	"torch" or "onnx"

DustyLM.chat

model.chat(
    messages: list[dict],
    temperature: float = 0.7,
    max_tokens: int = 64,
    top_p: float = 0.9,
) -> dict

Returns an OpenAI-style chat completion dict.

---

Built from the dusty-lm training repository.