## Context [WebTransport](https://www.w3.org/TR/webtransport/) extends HTTP/3 with multiplexed streams over QUIC. Lower latency than WebSocket, supports multiple parallel streams. ## Proposal - Use WebTransport to stream model layers in parallel from CDN - Multiple QUIC streams download different layer groups simultaneously - Priority scheduling: download layers needed soonest first - Combined with progressive inference (#383): start generating while layers still downloading - Graceful degradation to fetch() for browsers without WebTransport ## Impact - 2-4x faster model loading via parallel streams - Reduced time-to-first-token on slow connections
Context
WebTransport extends HTTP/3 with multiplexed streams over QUIC. Lower latency than WebSocket, supports multiple parallel streams.
Proposal
Impact