Repository navigation
- Fix split-device quantization issues
- Improved checkpointing
- Pipelined prefill
- Fix issues with disk tier shared across multiple instances
- More precise expert routing and sparse attn indexing
- Other optimizations & bugfixes
Full Changelog: v1.6.6...v1.6.7