Parti-Base

Turns a plain-English hardware idea into an organized build plan.

Tell it what to build — "a compact desk clock with an e-ink display and an IR remote" — and it returns one structured blueprint plan: the parts, how they connect, ordered build steps, rough sourcing and cost, and a quick design check. It's a standalone, all-in-one model (adapter-only version upon request).

Early research preview. For drafting and exploring ideas — not a replacement for real engineering, CAD, or safety review.

What it does

Give it a hardware idea and it returns, as one machine-readable JSON object, any of:

• 📋 a parts list
• 🔌 a wiring / connection map between the parts
• 🛠️ ordered build steps
• 💲 rough sourcing and cost
• ✅ a basic design check
• 📦 or the whole plan at once

Ask for the complete plan, or just one piece (like only the parts list).

Results

We test on projects it has never seen during training. How often it produces a valid, well-structured result for each task:

Task	Valid result
🛠️ Build steps	~100%
✅ Design check	~100%
📋 Parts list	~95%
📦 Full project plan	~85–97%
🔌 Wiring map	~67%

It's strongest at build steps, design checks, and parts lists; full end-to-end plans are close behind; wiring maps are the hardest. Figures are from held-out testing and are being finalized for the current version.

Because it's a small model, treat the output as a helpful first draft to review, not a finished design.

Where it fits

• Reliable structured JSON from plain English on a small (3B) model that runs on modest hardware.
• For images (sketches / renders), improved quality, and larger projects, use parti-vision (9B, multimodal).

What you can give it

• A plain-English request — one or two sentences. Text only

Try it

from transformers import AutoModelForCausalLM, AutoTokenizer

REPO = "caid-technologies/parti-base"
model = AutoModelForCausalLM.from_pretrained(REPO, device_map="auto", torch_dtype="bfloat16")
tok = AutoTokenizer.from_pretrained(REPO)

msgs = [
    {"role": "system", "content":
        "You design maker/electronics products. Given a request, reply with a single JSON object "
        "describing the complete build plan. Output only the JSON."},
    {"role": "user", "content": "A compact desk clock with an e-ink display and an IR remote."},
]
inputs = tok.apply_chat_template(
    msgs, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)
out = model.generate(**inputs, max_new_tokens=6144, do_sample=False,
                     repetition_penalty=1.1, pad_token_id=tok.eos_token_id)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

💡 Tips: keep do_sample=False (greedy — sampling degrades the JSON), keep max_new_tokens high (≥ 6000) so long plans aren't cut off, and keep repetition_penalty=1.1 so wiring lists don't get stuck repeating. For Ollama / local apps, convert to GGUF with llama.cpp.

Good to know

• It's a small model, so complex, many-part projects are harder for it.
• It proposes designs; it doesn't verify them. Always sanity-check before building.
• It's strongest on common project types (lab tools, smart-home) and weaker on rarer ones.

Learn more

• 📄 Technical whitepaper (PDF) — what it does, training approach, evaluation methodology, and per-task results.
• 💬 Discord community — questions, builds, feedback.
• 🚀 Want images and higher quality? See caid-technologies/parti-vision, the flagship 9B multimodal model.

Citation

@misc{parti_base,
  title  = {Parti-Base},
  author = {Caid Technologies},
  year   = {2026},
  howpublished = {\url{https://huggingface.co/caid-technologies}}
}