Lifecycle improvements: - runloop.go: refined state machine with better error recovery - service_manager.go: service dependency management and health checks - states.go: add states for capability advertisement and draining Container execution: - container.go: improved OCI runtime integration with supply chain checks - Add image verification and signature validation - Better resource limits enforcement for GPU/memory vLLM plugin updates: - vllm.go: support for vLLM 0.3+ with new engine arguments - Add quantization-aware scheduling (AWQ, GPTQ, FP8) - Improve model download and caching logic Configuration: - config.go: add capability advertisement configuration - snapshot_store.go: improve snapshot management for checkpointing |
||
|---|---|---|
| .. | ||
| jupyter.go | ||
| vllm.go | ||