What it does
Fork of vLLM optimized for Apple Silicon, focused on reliable serving of vision-language models locally. Prioritizes serving stability, compatibility evidence, and publishable benchmarks over speculative performance claims.
Key capabilities
- Local VLM serving on M-series hardware
- GUI-specialized benchmark suite for UI agent workloads
- Model compatibility matrix across hardware generations
- Client/app compatibility testing
Key numbers
- 12 models validated on M-series
- 6 VLMs tested for GUI-specific tasks
- 3 chip generations covered (M1, M2, M3+)
Current phase
MISSING — Current development priorities from roadmap
Status
Active. Evidence collection and benchmark methodology established.
Links
MISSING — Repository URL