RuntimeError: FlashAttention only support Ampere GPUs or newer despite setting VLLM_ATTENTION_BACKEND=XFORMERS

Description: I’m using:

vllm==0.6.3

GPU: Titan RTX (Turing architecture, not Ampere+)

Environment variable: export VLLM_ATTENTION_BACKEND=XFORMERS

When running bash ./scripts/train_tiny_zero.sh, I still encounter the error: “RuntimeError: FlashAttention only supports Ampere GPUs or newer. ”

What would be the recommended solution for this? Thanks for your help.