Skip to content

v1.1.0

Choose a tag to compare

View all tags
@jjang-ai jjang-ai released this 06 Mar 04:03
· 4127 commits to main since this release
v1.1.0
1445e59

What's New

New Features

  • Speculative Decoding UI — Configure draft model and num-draft-tokens for 20-90% speedup
  • Generation Defaults — Server-wide default temperature and top-p sliders
  • Embedding Model — Pre-load a separate embedding model for /v1/embeddings endpoint
  • Auto-Update Checker — Dismissible banner notification when a new version is available

Fixes

  • VLM Batching — Fixed !isVLM guard that prevented VLMs from using BatchedEngine with MLLMScheduler
  • Parser Priority — User explicit choice now correctly wins over auto-detection in both buildArgs() and preview

Tests

  • 95 settings-flow tests (was 55), covering all new fields, parser priority, VLM batching
  • 447 total tests passing across 6 test files
Assets 2
Loading