perf(qwen27 cuda): reduce TTFT with prompt caching and state reuse - #2

perf(qwen27 cuda): reduce TTFT with prompt caching and state reuse#2

jregan-beasley-smc wants to merge 4 commits into

machinegenieorg:see-4347-qwen27-cudamachinegenieorg/TensorFold:see-4347-qwen27-cudafrom

jregan-beasley-smc:perf/qwen27-exact-prompt-cachejregan-beasley-smc/TensorFold:perf/qwen27-exact-prompt-cacheCopy head branch name to clipboard