20261026_Krea2Edit_omnixpu加速_B70_64G_冷启+热启.txt · 2026-10-26 22:53
| 阶段 | 耗时 | 步数 | s/it | staged MB | 其他指标 |
|---|---|---|---|---|---|
| 冷启 | 115.97 s | 8 | 2.05 | 12,864 | 0.1 MB freedfallback ×5CUTE ×3 |
| 热启 | 36.96 s | 8 | 1.79 | 12,864 | 0.1 MB freed |
[INFO] got prompt [INFO] Model storage policy: fast_disk=False paths=['<ComfyUI>/models/vae/qwen_image_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.float32 # 😺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=['<ComfyUI>/models/text_encoders/qwen3vl_4b_fp8_scaled.safetensors'] [INFO] CLIP/text encoder model load device: xpu:0, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] Requested to load Krea2TEModel_ [INFO] Model Krea2TEModel_ prepared for dynamic VRAM loading. 4999MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB. [INFO] [OmniXPU] norm first use: LayerNorm shape=torch.Size([1728, 1024]) [INFO] Model Krea2TEModel_ prepared for dynamic VRAM loading. 4999MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=4406.2MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=4500.0MiB [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: asym_w4a8_int8, float8_e5m2, nvfp4, w6a8_int8, mxfp8, float8_e4m3fn, convrot_w4a4 [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>/krea2_turbo_int8_convrot.safetensors'] [INFO] [OmniXPU] LoRA memory budget cached: source=1743.5MiB/512 tensors model_added=1743.5MiB model_budget=1743.5MiB clip_added=0.0MiB clip_budget=0.0MiB [krea2edit] pre-encoding sources at target 1024x768px (before sampling, fit_mode=fit) [krea2edit] _fit_encode_image: mode=fit in=(1, 1024, 768, 3) target_latent=128x96 [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=4500.0MiB [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [krea2edit] _fit_encode_image: mode=fit in=(1, 752, 1024, 3) target_latent=128x96 [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2460.9MiB [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] [OmniXPU] LoRA model budget applied: cached=1743.5MiB/512 tensors requested=1081.3->2824.8MiB minimum=540.7->2284.2MiB xpu_memory=available xpu_allocated=0.7MiB xpu_reserved=2.0MiB xpu_free=23299.2MiB xpu_total=31977.4MiB [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2284.2MiB [INFO] Requested to load Krea2 [INFO] Model Krea2 prepared for dynamic VRAM loading. 12864MB Staged. 256 patches attached. Force pre-loaded 160 weights: 2824 KB. [INFO] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=3.5MiB xpu_reserved=4.0MiB xpu_free=23294.7MiB xpu_total=31977.4MiB 0%| | 0/8 [00:00<?, ?it/s, Model Initializing ... ][krea2edit] pixel path ACTIVE (fit_mode=fit) [INFO] [OmniXPU] attention fallback: batch=891 (seq=12) [INFO] [OmniXPU] attention fallback: batch=891 (seq=12) [INFO] [OmniXPU] attention CUTE #1: heads=20 seq=891 dtype=torch.bfloat16 [INFO] [OmniXPU] attention CUTE #2: heads=20 seq=891 dtype=torch.bfloat16 [krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) [INFO] [OmniXPU] attention CUTE #3: heads=48 seq=8715 dtype=torch.bfloat16 12%|██████▏ | 1/8 [01:05<07:39, 65.65s/it, Model Initialization complete! ][INFO] [OmniXPU] attention fallback: batch=891 (seq=12) [INFO] [OmniXPU] attention fallback: batch=891 (seq=12) [krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) 25%|█████████████████████ | 2/8 [00:04<00:12, 2.05s/it][INFO] [OmniXPU] attention fallback: batch=891 (seq=12) [krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) 38%|███████████████████████████████▌ | 3/8 [00:06<00:10, 2.04s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) 50%|██████████████████████████████████████████ | 4/8 [00:08<00:08, 2.04s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) 62%|████████████████████████████████████████████████████▌ | 5/8 [00:10<00:06, 2.04s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) 75%|███████████████████████████████████████████████████████████████ | 6/8 [00:12<00:04, 2.06s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) 88%|█████████████████████████████████████████████████████████████████████████▌ | 7/8 [00:14<00:02, 2.04s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 48), (35, 48)] centered in (64,48) 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:16<00:00, 2.05s/it] VRAM清理完成 [卸载模型: True, 清空缓存: True] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=550.0MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] VRAMdebug: free memory before: 33,530,239,488 [INFO] VRAMdebug: free memory after: 33,530,337,792 [INFO] VRAMdebug: freed memory: 98,304 [INFO] Prompt executed in 115.97 seconds [INFO] Using RAM pressure cache. [INFO] got prompt [INFO] Model storage policy: fast_disk=False paths=['<ComfyUI>/models/vae/qwen_image_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.float32 # 😺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=['<ComfyUI>/models/text_encoders/qwen3vl_4b_fp8_scaled.safetensors'] [INFO] CLIP/text encoder model load device: xpu:0, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] Requested to load Krea2TEModel_ [INFO] Model Krea2TEModel_ prepared for dynamic VRAM loading. 4999MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB. [INFO] Model Krea2TEModel_ prepared for dynamic VRAM loading. 4999MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=4406.2MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=4125.0MiB [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: asym_w4a8_int8, float8_e5m2, nvfp4, w6a8_int8, mxfp8, float8_e4m3fn, convrot_w4a4 [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>/krea2_turbo_int8_convrot.safetensors'] [INFO] [OmniXPU] LoRA memory budget cached: source=1743.5MiB/512 tensors model_added=1743.5MiB model_budget=1743.5MiB clip_added=0.0MiB clip_budget=0.0MiB [krea2edit] pre-encoding sources at target 1024x704px (before sampling, fit_mode=fit) [krea2edit] _fit_encode_image: mode=fit in=(1, 1024, 704, 3) target_latent=128x88 [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=4125.0MiB [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [krea2edit] _fit_encode_image: mode=fit in=(1, 752, 1024, 3) target_latent=128x88 [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2062.5MiB [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] [OmniXPU] LoRA model budget applied: cached=1743.5MiB/512 tensors requested=991.2->2734.7MiB minimum=495.6->2239.1MiB xpu_memory=available xpu_allocated=1.1MiB xpu_reserved=4.0MiB xpu_free=22270.9MiB xpu_total=31977.4MiB [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=2239.1MiB [INFO] Requested to load Krea2 [INFO] Model Krea2 prepared for dynamic VRAM loading. 12864MB Staged. 256 patches attached. Force pre-loaded 160 weights: 2824 KB. [INFO] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=3.9MiB xpu_reserved=4.0MiB xpu_free=22270.9MiB xpu_total=31977.4MiB 0%| | 0/8 [00:00<?, ?it/s, Model Initializing ... ][krea2edit] pixel path ACTIVE (fit_mode=fit) [krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 12%|██████████████▌ | 1/8 [00:09<01:08, 9.85s/it, Model Initialization complete! ][krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 25%|█████████████████████████████████████▊ | 2/8 [00:03<00:10, 1.78s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 38%|████████████████████████████████████████████████████████▋ | 3/8 [00:05<00:08, 1.79s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 50%|███████████████████████████████████████████████████████████████████████████▌ | 4/8 [00:07<00:07, 1.78s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 62%|██████████████████████████████████████████████████████████████████████████████████████████████▍ | 5/8 [00:08<00:05, 1.79s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 75%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████▎ | 6/8 [00:10<00:03, 1.80s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 88%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▏ | 7/8 [00:12<00:01, 1.79s/it][krea2edit] STRIDE1-POS fit: ref grids [(64, 44), (32, 44)] centered in (64,44) 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:14<00:00, 1.79s/it] VRAM清理完成 [卸载模型: True, 清空缓存: True] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=550.0MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 483MB Staged. 0 patches attached. Force pre-loaded 60 weights: 89 KB. [INFO] VRAMdebug: free memory before: 33,530,239,488 [INFO] VRAMdebug: free memory after: 33,530,337,792 [INFO] VRAMdebug: freed memory: 98,304 [INFO] Prompt executed in 36.96 seconds [INFO] Using RAM pressure cache.