Skip to main content

Ultimate AI Connector for WebLLM Ilogu yezinguquko

Inguqulo 1.1.0 — Ikhishwe ngo-2026-04-09​

  • Okusha: Iwijethi yengxoxo entantayo enenkomba yesimo ye-admin-bar — noma yimuphi umsebenzisi ongene ngemvume angathumela umyalo ku-LLM eseceleni kwesiphequluli ngqo kusuka ku-front end.
  • Okusha: Isikhathi sokusebenza se-SharedWorker — injini ye-MLC manje isebenza ku-SharedWorker ukuze amathebhu amaningi abelane ngeseshini eyodwa ye-GPU esikhundleni sokuyilwela.
  • Okusha: Isibambi esiphakathi se-apiFetch — izicelo ze-WordPress REST ezihambisana nephethini ye-AI Client SDK ziqondiswa ngokusobala ku-broker yasendaweni ye-WebLLM, akukho ukujikeleza kwe-loopback HTTP okudingekayo.
  • Okusha: I-UI yezilungiselelo zewijethi kuphaneli ye-Connector yokuvula noma ukuvala iwijeithi yengxoxo nokulungiselela ukuziphatha kwe-auto-prompt.
  • Ukulungisa: Phoqa i-backend yenqolobane ye-IndexedDB ukuze ukulandwa kwesisindo semodeli kusinde ekuqondisweni kabusha kwe-HuggingFace xet CDN okuphula indlela ezenzakalelayo ye-Cache API.
  • Ukulungisa: Yeqa ukushintshwa kwe-context_window KV-cache kumamodeli e-embedding (awanayo i-decoder futhi ukushintsha kwakudala kubangele iphutha lesikhathi sokusebenza).
  • Ukulungisa: Memezela imodeli yekhandidethi ye-cold-start ku-/webllm/v1/models ngaphambi kokuba ithebhu yesisebenzi ilayishe, ukuze abathengi be-SDK babone imodeli ngokushesha.
  • Ukulungisa: Xhuma i-WebLLM kusihlungi se-wpai_preferred_text_models ukuze isici se-AI Experiments siqondise enjinini yesiphequluli lapho imodeli ilungiselelwe.
  • Okuthuthukisiwe: Ukugwema inqolobane, ukujwayelekisa okuqukethwe, nokulungiswa kwezinkomba zehadiwe okuvele ngesikhathi sokuhlola kusukela ekuqaleni kuya ekugcineni.

Inguqulo 1.0.0 — Ukukhishwa kokuqala​