electron-rare/mascarade-esp32

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.1BQuant:BF16Context Size:2kPublished:Mar 9, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

electron-rare/mascarade-esp32 is a 1.1 billion parameter TinyLlama-based model fine-tuned by electron-rare, specifically optimized for ESP32 microcontroller development. This model excels at generating code and providing technical assistance related to ESP32 and ESP-IDF firmware. It is part of the Mascarade ecosystem, focusing on domain-specific applications for embedded systems.

Loading preview...

Mascarade ESP32: Specialized LLM for ESP32 Development

electron-rare/mascarade-esp32 is a 1.1 billion parameter language model, fine-tuned from TinyLlama/TinyLlama-1.1B-Chat-v1.0, with a 2048-token context length. Developed by electron-rare, this model is specifically designed to assist with ESP32 microcontroller development and is part of the broader Mascarade ecosystem for embedded systems.

Key Capabilities

  • ESP32-specific knowledge: Highly specialized in generating code and providing information for ESP32 and ESP-IDF firmware.
  • Efficient size: Built on TinyLlama, offering a compact 1.1B parameters suitable for resource-conscious applications.
  • LoRA fine-tuning: Trained using LoRA (PEFT) with a domain-specific dataset in ShareGPT format, focusing on ESP32 examples.

Training Details

The model underwent 2 epochs of training over 30 steps, achieving a final training loss of 1.3873. The LoRA configuration included a rank of 16, alpha of 32, and a dropout of 0.05, targeting q_proj, k_proj, v_proj, and o_proj modules. The training utilized a Quadro P2000 GPU and the Hugging Face Transformers + PEFT framework.

Intended Use

This model is intended for use in ESP32 / ESP-IDF firmware development. It is explicitly out of scope for critical applications such as healthcare diagnosis, legal advice, autonomous safety-critical decisions, or the generation of malicious code. Users are advised to independently verify all outputs, as AI models may produce inaccurate or biased information.