FLUX.2 [klein] 4B×

Resolution: 1024×1024
Batch: 1
Steps: 50
Precision: BF16
Guidance: 4
Attention: torch_sdpa
Offload: none
Protocol: v0.1
Prompts: 20 public
Updated: 2026-10-01

Drag the dashed line (or focus its grip and use the arrow keys) to set the quality limit. The fastest recipe within it is selected; recipes above it are greyed out.

Fastest within LPIPS ≤ .050

Baseline17.8s · 1.0× faster · LPIPS —

Latency	LPIPS	Speedup	Recipe	Engine	Verified
2.6s	.193	6.9×	FP8 W8A8 + SageAttention2 + fused concat + FP8 quantize + DPCache K=16	sglang-diffusion 5ead00ae6	✓
3.1s	.120	5.8×	FP8 W8A8 (double-stream blocks BF16) + SageAttention2 + fused concat + FP8 quantize + DPCache K=16	sglang-diffusion 5ead00ae6	✓
8.0s	.188	2.2×	FP8 W8A8 + SageAttention2	sglang-diffusion 5ead00ae6	✓
17.8s	—	1.0×	Baseline	sglang-diffusion 5ead00ae6	✓

Add a recipe

Found a better optimization combination? Submit a reproducible recipe through GitHub.

Submit via GitHub ↗

Submissions are evaluated under the same model, hardware, workload, and quality protocol. The submission unit is a complete recipe.