Pockot珀刻機

工具 v1

可行性測算器

啟發式可行性模型。它估算續航、記憶體適配與粗略假設層級；該層級不是能力保證。

Corpus profileEnglish Wikipedia all nopic ZIM, 2026-0652.7 GB · Wikimedia index lists 52,690,706,555 bytes; excludes pictures.

wikimedia-kiwix-zim

Corpus profileEnglish Wikipedia medicine maxi ZIM, 2026-042.3 GB · Wikimedia index lists 2,215,520,569 bytes; useful as a small medical reference pack.

wikimedia-kiwix-zim

Corpus profileCurated field manual pack8 GB · Pockot planning placeholder for user-approved manuals and notes; source/date registry required.

wikimedia-kiwix-zim

Device scaffoldJetson field reference scaffold$249 · Board plus battery boundary only; storage, enclosure, cooling, display, and measured runtime are not included.

nvidia-jetson-orin-nano-super, faa-batteries

Device scaffoldRaspberry Pi AI HAT+ 2 scaffold$200 · AI HAT+ 2 price and accelerator specs only; Raspberry Pi 5 host, storage, enclosure, and measured runtime are separate.

raspberry-pi-ai-hat-2, raspberry-pi-ai-hat, faa-batteries

Adaptation modeRetrieval-only updatesUpdates source packs and indexes without changing model weights.

nist-ai-rmf, owasp-llm-top10

Adaptation modeAdapter experimentParameter-efficient training must keep evaluation, rollback, and data-quality checks visible.

lora, qlora

啟發式可行性模型。它估算續航、記憶體適配與粗略假設層級；該層級不是能力保證。

來源

  1. Apple introduces M4 chip
  2. Llama 3.2 edge and mobile devices
  3. Raspberry Pi AI HAT+ documentation
  4. Raspberry Pi AI HAT+ 2
  5. NVIDIA Jetson Orin Nano Super Developer Kit
  6. Wikimedia Kiwix ZIM index
  7. FAA airline passengers and batteries
  8. Snapdragon X Elite product specifications
  9. LoRA: Low-Rank Adaptation of Large Language Models
  10. QLoRA: Efficient Finetuning of Quantized LLMs
  11. NREL interactive best research-cell efficiency chart
  12. NIST AI Risk Management Framework
  13. OWASP Top 10 for Large Language Model Applications