Release 2.49.0 corresponding to NGC container 24.08

What's Changed

• refactor: Remove explicit callings to garbage collect by @kthui in #55
• perf: Check for cancellation on response thread by @kthui in #54
• feat: Add vLLM counter metrics access through Triton by @yinggeh in #53
• feat: Report histogram metrics to Triton metrics server by @yinggeh in #58
• feat: Report more histogram metrics by @yinggeh in #61

Full Changelog: v24.07...v24.08

Release 2.48.0 corresponding to NGC container 24.07

What's Changed

• Removed explicit mode for multi-lora by @oandreeva-nv in #45
• test: Limiting multi-gpu tests to use Ray as distributed_executor_backend by @oandreeva-nv in #47
• perf: Improve vLLM backend performance by using a separate thread for responses by @Tabrizian in #46

Full Changelog: v24.06...v24.07

Release 2.47.0 corresponding to NGC container 24.06

fix: Enhance checks around KIND_GPU and tensor parallelism (#42)

Co-authored-by: Olga Andreeva <[email protected]>