vLLM 0.32 changed four contracts at once
The release improves routing and memory, but it also changes browser policy, token admission, stream events, and runtime dependencies. Promote it as a serving migration, not a package update.
Topic · 4 articles
The latest analysis on Agent Operations, across Daily Pulses and Weekly Reviews.
The release improves routing and memory, but it also changes browser policy, token admission, stream events, and runtime dependencies. Promote it as a serving migration, not a package update.
LlamaIndex's long-tail extraction benchmark makes whole-document success, grounding, latency, and cost visible. The transferable lesson is to promote an extractor only after it survives production-shaped failure classes.
NVIDIA and Princeton report faster hybrid-model decoding by checkpointing recurrent state and replaying recent inputs. The deployment decision turns on batch size, rollback, and precision, not the headline speedup.
Cowork's Pro and Max cloud migration changes where code runs, where files are processed, and which controls remain on the device. Treat execution locality as governed configuration, not product trivia.