202610282123_Flux3Edit_omnixpu加速_B70_64G_冷启+热启.txt · 2026-10-28 21:23
| 阶段 | 耗时 | 步数 | s/it | staged MB | 其他指标 |
|---|---|---|---|---|---|
| 冷启 | 70.52 s | 16 | 0.77 | 5,557 | CUTE ×3 |
| 热启 | 32.22 s | 16 | 1.00 | 5,557 | — |
[INFO] got prompt [INFO] Model storage policy: fast_disk=False paths=['<ComfyUI>/models/vae/flux2-vae.safetensors'] [INFO] Using pytorch attention in VAE [INFO] Using pytorch attention in VAE [INFO] VAE load device: xpu:0, offload device: cpu, dtype: torch.bfloat16 # 😺dzNodes: LayerStyle -> ImageScaleByAspectRatio V2 Processed 1 image(s). # 😺dzNodes: LayerStyle -> ImageScaleByAspectRatio V2 Processed 1 image(s). [INFO] Found quantization metadata version 1 [INFO] Using MixedPrecisionOps for text encoder [INFO] Model storage policy: fast_disk=False paths=['<本地路径>/qwen3vl_4b_int8_convrot.safetensors'] [INFO] CLIP/text encoder model load device: xpu:0, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] Requested to load Flux3TEModel_ [INFO] Model Flux3TEModel_ prepared for dynamic VRAM loading. 4632MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB. [INFO] [OmniXPU] norm first use: rms_norm_fn shape=torch.Size([1, 80, 2560]) [INFO] Model Flux3TEModel_ prepared for dynamic VRAM loading. 4632MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2429.6MiB [INFO] Requested to load AutoencoderKL [INFO] Model AutoencoderKL prepared for dynamic VRAM loading. 160MB Staged. 0 patches attached. Force pre-loaded 112 weights: 101 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2595.3MiB [INFO] Model AutoencoderKL prepared for dynamic VRAM loading. 160MB Staged. 0 patches attached. Force pre-loaded 112 weights: 101 KB. [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: w6a8_int8, convrot_w4a4, asym_w4a8_int8, float8_e4m3fn, mxfp8, nvfp4, float8_e5m2 [INFO] model weight dtype torch.bfloat16, manual cast: torch.bfloat16 [INFO] model_type FLUX [INFO] Model storage policy: fast_disk=False paths=['<本地路径>/flux3-image-klein-int8-convrot.safetensors'] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2527.8MiB [INFO] Requested to load Flux3 [INFO] Model Flux3 prepared for dynamic VRAM loading. 5557MB Staged. 0 patches attached. Force pre-loaded 86 weights: 60 KB. 0%| | 0/16 [00:00<?, ?it/s, Model Initializing ... ] [INFO] [OmniXPU] attention CUTE #1: heads=24 seq=80 dtype=torch.bfloat16 [INFO] [OmniXPU] attention CUTE #2: heads=24 seq=80 dtype=torch.bfloat16 [INFO] [OmniXPU] attention CUTE #3: heads=24 seq=80 dtype=torch.bfloat16 0%| | 0/16 [00:23<?, ?it/s, Model Initialization complete!] 6%|███ | 1/16 [00:23<05:54, 23.61s/it, Model Initialization complete!] 100%|██████████████████████████████████████████████████████████████████████████████████| 16/16 [00:12<00:00, 1.30it/s] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=272.2MiB [INFO] Model AutoencoderKL prepared for dynamic VRAM loading. 160MB Staged. 0 patches attached. Force pre-loaded 112 weights: 101 KB. [INFO] Prompt executed in 70.52 seconds [INFO] got prompt # 😺dzNodes: LayerStyle -> ImageScaleByAspectRatio V2 Processed 1 image(s). [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=3037.0MiB [INFO] Model AutoencoderKL prepared for dynamic VRAM loading. 160MB Staged. 0 patches attached. Force pre-loaded 112 weights: 101 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2595.3MiB [INFO] Model AutoencoderKL prepared for dynamic VRAM loading. 160MB Staged. 0 patches attached. Force pre-loaded 112 weights: 101 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2939.8MiB [INFO] Model Flux3 prepared for dynamic VRAM loading. 5557MB Staged. 0 patches attached. Force pre-loaded 86 weights: 60 KB. 100%|█████████████████████████████████████████████████████████████████████████████████| 16/16 [00:15<00:00, 1.00it/s] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=272.2MiB [INFO] Model AutoencoderKL prepared for dynamic VRAM loading. 160MB Staged. 0 patches attached. Force pre-loaded 112 weights: 101 KB. [INFO] Prompt executed in 32.22 seconds