clemsail/mascarade-iot

Mascarade IoT

Fine-tuned Qwen2.5-Coder-1.5B-Instruct model specialized in IoT (Internet of Things) for embedded electronics.

Part of the Mascarade ecosystem — an agentic LLM orchestration system with domain-specific fine-tuned models for embedded systems and electronics.

Training details

Parameter	Value
Base model	Qwen/Qwen2.5-Coder-1.5B-Instruct
Method	LoRA (PEFT) — merged into full weights
LoRA rank (r)	16
LoRA alpha	32
LoRA dropout	0.05
Target modules	q_proj, k_proj, v_proj, o_proj
Epochs	~2
Training steps	58
Final train loss	1.2653
Dataset	clemsail/mascarade-iot-dataset (ShareGPT format)
GPU	Quadro P2000 (5 GB VRAM)
Framework	Hugging Face Transformers + PEFT

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("electron-rare/mascarade-iot")
tokenizer = AutoTokenizer.from_pretrained("electron-rare/mascarade-iot")

messages = [{"role": "user", "content": "How do I set up MQTT on an ESP32?"}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt")
outputs = model.generate(inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Related models

Model	Domain	Base
mascarade-esp32	ESP32 microcontrollers	TinyLlama-1.1B
mascarade-spice	SPICE circuit simulation	TinyLlama-1.1B
mascarade-platformio	PlatformIO development	TinyLlama-1.1B

Datasets

All training datasets are available under clemsail on Hugging Face.