Release candidate #2 for MLPerf Inference v6.0

What's Changed

• Add bilinear_pos_embed triton kernel and cache by @zhandaz in #40
• Support Flashinfer FP8 ViT Attn by @zhandaz in #41
• Nvimgcodec + tensor ipc by @soodoshll in #38
• Add nvjpeg into requirements/cuda.txt by @wangshangsam in #42
• Revert "Nvimgcodec + tensor ipc (#38)" and "Add nvjpeg into requirements/cuda.txt (#42)" by @wangshangsam in #43
• Pin FlashInfer version to RC2 by @wangshangsam in #44

Full Changelog: mlperf-inf-mm-q3vl-v6.0-rc1...mlperf-inf-mm-q3vl-v6.0-rc2

Release candidate #1 for MLPerf Inference v6.0

What's Changed

• Fix TRTLLM attention assert by @soodoshll in #28
• Revert "[Improvement] Persist CUDA compat libraries paths to prevent reset on apt-get (vllm-project#30784)" by @wangshangsam in #31
• FA4 attention for ViT by @zhandaz in #29
• Add Flashinfer cudnn Backend for ViT by @maxyanghu in #30
• Shortest job first scheduling for server mode by @soodoshll in #32
• Remove flashinfer show-config from Dockerfile by @wangshangsam in #34
• Pin FlashInfer version by @wangshangsam in #39

New Contributors

• @soodoshll made their first contribution in #28
• @wangshangsam made their first contribution in #31
• @zhandaz made their first contribution in #29
• @maxyanghu made their first contribution in #30

Full Changelog: https://github.com/CentML/vllm/commits/mlperf-inf-mm-q3vl-v6.0-rc1