caid-technologies/parti-base

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 21, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Parti-Base by Caid Technologies is a 3 billion parameter language model based on Qwen2.5-3B, specifically designed to convert plain-English hardware ideas into structured build plans. It excels at generating machine-readable JSON outputs for parts lists, build steps, and design checks for maker/electronics projects. This model is optimized for structured output generation on modest hardware, providing a helpful first draft for hardware design and exploration.

Loading preview...

Overview

Parti-Base, developed by Caid Technologies, is a specialized 3 billion parameter language model built upon the Qwen2.5-3B architecture. Its core function is to transform plain-English descriptions of hardware ideas into organized, machine-readable JSON build plans. This model is particularly adept at generating structured outputs for maker and electronics projects, making it suitable for drafting and exploring hardware concepts.

Key Capabilities

  • Structured JSON Output: Generates comprehensive build plans, parts lists, wiring maps, build steps, sourcing/cost estimates, and basic design checks as a single JSON object.
  • Hardware Idea Generation: Translates natural language requests (e.g., "a compact desk clock with an e-ink display") into actionable project blueprints.
  • Modest Hardware Compatibility: Designed to run efficiently on less powerful hardware due to its small size (3B parameters).
  • High Reliability for Specific Tasks: Achieves near 100% valid results for build steps and design checks, and 85-97% for full project plans.

When to Use Parti-Base

  • Drafting Hardware Projects: Ideal for generating initial drafts and exploring ideas for maker/electronics projects.
  • Structured Data Needs: When reliable, structured JSON output from natural language input is required.
  • Resource-Constrained Environments: Suitable for applications where a smaller model footprint is necessary.

Limitations

  • Early Research Preview: Outputs are intended as a first draft and require review; not a replacement for professional engineering or safety analysis.
  • Complexity: More complex or multi-part projects may be challenging for this small model.
  • Verification: The model proposes designs but does not verify them; human sanity checks are essential.
  • Wiring Maps: This is the hardest task for the model, with approximately 67% valid results.

For projects requiring images, improved quality, or larger scope, consider caid-technologies/parti-vision, a 9B multimodal model.

Finetunes
1 models