[gemma4_31b][cuda] length-aware bf16 global attention #7282

Job	Run time
Get changed files / get-changed-files	8s
CI run decision / decide	34s
export-model-cuda-windows-artifact (mistralai, Voxtral-Mini-3B-2507, non-quantized) / linux-job	28m 55s
export-model-cuda-windows-artifact (facebook, dinov2-small-imagenet1k-1-layer, non-quantized) / linux-job	15m 45s
export-model-cuda-windows-artifact (mistralai, Voxtral-Mini-3B-2507, quantized-int4-weight-only) / linux-job	29m 50s
export-model-cuda-windows-artifact (nvidia, parakeet-tdt, quantized-int4-weight-only) / linux-job	25m 33s
export-model-cuda-windows-artifact (mistralai, Voxtral-Mini-4B-Realtime-2602, quantized-int4-tile... / linux-job	29m 29s
export-model-cuda-windows-artifact (nvidia, parakeet-tdt, non-quantized) / linux-job	25m 20s
test-model-cuda-windows-e2e (facebook, dinov2-small-imagenet1k-1-layer, non-quantized) / windows-job	33m 29s
test-model-cuda-windows-e2e (mistralai, Voxtral-Mini-4B-Realtime-2602, quantized-int4-tile-packed) / windows-job	33m 57s
test-model-cuda-windows-e2e (mistralai, Voxtral-Mini-3B-2507, non-quantized) / windows-job	36m 38s
test-model-cuda-windows-e2e (mistralai, Voxtral-Mini-3B-2507, quantized-int4-weight-only) / windows-job	35m 30s
test-model-cuda-windows-e2e (nvidia, parakeet-tdt, non-quantized) / windows-job	35m 16s
test-model-cuda-windows-e2e (nvidia, parakeet-tdt, quantized-int4-weight-only) / windows-job	34m 11s
	6h 4m 35s

Provide feedback