Release: v0.17.2
What's Changed
- Prevented embedding batches from stalling on documents containing unusually long inputs.
- Added token-aware outlier deferral to process regular batches first and oversized inputs individually.
- Preserved original chunk content, embedding order, and model-specific token limits.
Full Changelog: v0.17.1...v0.17.2