Skip to main content

Ultimate AI Connector for WebLLM Ingxelo yotshintsho

Inguqulelo 1.1.0 — Ikhutshwe ngo-2026-04-09​

  • Entsha: Iwijethi yencoko edadayo enesalathisi sobume kwi-admin-bar — nawuphi na umsebenzisi ongene ngemvume angayalela i-LLM esecaleni lesikhangeli ngokuthe ngqo ukusuka kwicala langaphambili.
  • Entsha: Ixesha lokusebenza le-SharedWorker — injini ye-MLC ngoku isebenza kwi-SharedWorker ukuze iithebhu ezininzi zabelane ngeseshoni enye ye-GPU endaweni yokulwela yona.
  • Entsha: Umngeneleli we-apiFetch middleware — izicelo ze-WordPress REST ezihambelana nepateni ye-AI Client SDK zihanjiswa ngokungafihlisiyo kwi-broker yasekuhlaleni ye-WebLLM, akukho mfuneko yohambo olubuyela kwi-HTTP ye-loopback.
  • Entsha: Ujongano lweesetingi zewijethi kwiphaneli ye-Connector lokuvula okanye lokuvala iwijethi yencoko nokumisela indlela yokuziphatha kwe-auto-prompt.
  • Ulungiso: Nyanzelisa i-cache backend ye-IndexedDB ukuze ukhuphelo lobunzima bemodeli lusinde kwi-HuggingFace xet CDN redirects ezaphula indlela engagqibekanga ye-Cache API.
  • Ulungiso: Tsiba i-context_window KV-cache override kwiimodeli ze-embedding (azinayo i-decoder kwaye i-override ibangele impazamo yexesha lokusebenza).
  • Ulungiso: Yazisa imodeli engumgqatswa we-cold-start ku-/webllm/v1/models ngaphambi kokuba ithebhu ye-worker ilayishe, ukuze abasebenzisi be-SDK babone imodeli ngoko nangoko.
  • Ulungiso: Qhagamshela i-WebLLM kwi-wpai_preferred_text_models filter ukuze uphawu lwe-AI Experiments luhambise kwinjini yesikhangeli xa imodeli imiselwe.
  • Kuphuculwe: Ukuphelisa i-cache, ukulinganisa umxholo, kunye nolungiso lweereferensi zehardware ezivele ngexesha lovavanyo lokuphela ukuya ekupheleni.

Inguqulelo 1.0.0 — Ukukhutshwa kokuqala​