[None][feat] DFlash: TRTLLM drafter attention by default wherever it runs, cold-page compression with a TRTLLM drafter, and acceptance-length tests - #21535

[None][feat] DFlash: TRTLLM drafter attention by default wherever it runs, cold-page compression with a TRTLLM drafter, and acceptance-length tests#21535

zheyuf wants to merge 5 commits into

NVIDIA:feat/musespark-open-agentxNVIDIA/TensorRT-LLM:feat/musespark-open-agentxfrom

zheyuf:zheyuf/dflash-trtllm-default-al-testszheyuf/TensorRT-LLM:zheyuf/dflash-trtllm-default-al-testsCopy head branch name to clipboard