[WS1][CUDA][Qwen3-Next] Full-model chain: vLLM rollout vs VIME recompute, raw-bit through two updates and a reload (RFC #428 stack 7/8) - #543

[WS1][CUDA][Qwen3-Next] Full-model chain: vLLM rollout vs VIME recompute, raw-bit through two updates and a reload (RFC #428 stack 7/8)#543

fusheng-ji wants to merge 81 commits into

RL-Align:test-qwennextRL-Align/RL-Kernel:test-qwennextfrom

fusheng-ji:feat/ws1-qwen3-next-full-model-chainfusheng-ji/RL-Kernel:feat/ws1-qwen3-next-full-model-chainCopy head branch name to clipboard