202610271640_Qwenimage2.1Edit_omnixpu加速_B70_64G_冷启+热启.txt · 2026-10-27 16:40
| 阶段 | 耗时 | 步数 | s/it | staged MB | 其他指标 |
|---|---|---|---|---|---|
| 冷启 | 99.59 s | 8 | 0.57 | 8,916 | 0.1 MB freedfallback ×5CUTE_BHLD_D128 ×3 |
| 热启 | 20.88 s | 8 | 0.52 | 8,916 | 0.1 MB freed |
[INFO] got prompt [INFO] Model storage policy: fast_disk=False paths=['<ComfyUI>/models/vae/qwen_image_2.1_vae_bf16.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 [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_8b_int8_convrot.safetensors'] [INFO] CLIP/text encoder model load device: xpu:0, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=1214.1MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 644MB Staged. 0 patches attached. Force pre-loaded 68 weights: 102 KB. [INFO] Requested to load QwenImage21TEModel_ [INFO] Model QwenImage21TEModel_ prepared for dynamic VRAM loading. 8916MB Staged. 0 patches attached. Force pre-loaded 255 weights: 849 KB. [INFO] [OmniXPU] norm first use: LayerNorm shape=torch.Size([4144, 1152]) [INFO] Model QwenImage21TEModel_ prepared for dynamic VRAM loading. 8916MB Staged. 0 patches attached. Force pre-loaded 255 weights: 849 KB. [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: nvfp4, float8_e4m3fn, float8_e5m2, mxfp8, w6a8_int8, asym_w4a8_int8, 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>/qwen_image_2.1_int8_convrot.safetensors'] [INFO] [OmniXPU] LoRA memory budget cached: source=152.0MiB/384 tensors model_added=152.0MiB model_budget=152.0MiB clip_added=0.0MiB clip_budget=0.0MiB [INFO] [OmniXPU] LoRA model budget applied: cached=152.0MiB/384 tensors requested=1971.4->2123.4MiB minimum=985.7->1137.7MiB xpu_memory=available xpu_allocated=1.5MiB xpu_reserved=4.0MiB xpu_free=20161.8MiB xpu_total=31977.4MiB [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=1137.7MiB [INFO] Requested to load QwenImage21 [INFO] Model QwenImage21 prepared for dynamic VRAM loading. 6920MB Staged. 192 patches attached. Force pre-loaded 65 weights: 63 KB. [INFO] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=1.6MiB xpu_reserved=4.0MiB xpu_free=20193.5MiB xpu_total=31977.4MiB 0%| | 0/8 [00:00<?, ?it/s, Model Initializing ... ][INFO] [OmniXPU] attention fallback: mask=torch.Size([8, 8]) (seq=8) [WARNING] [OmniXPU] experimental capability-driven BMG D128 CUTE route enabled; set OMNI_ATTN_BACKEND=torch to disable it if this workflow shows an issue [INFO] [OmniXPU] attention CUTE_BHLD_D128 #1: heads=32 q=4144 kv=4152 dtype=torch.bfloat16 [INFO] [OmniXPU] attention fallback: mask=torch.Size([64, 4216]) (seq=64) [INFO] [OmniXPU] attention CUTE_BHLD_D128 #2: heads=32 q=4070 kv=8286 dtype=torch.bfloat16 [INFO] [OmniXPU] attention fallback: mask=torch.Size([8, 8]) (seq=8) [INFO] [OmniXPU] attention CUTE_BHLD_D128 #3: heads=32 q=4144 kv=4152 dtype=torch.bfloat16 [INFO] [OmniXPU] attention fallback: mask=torch.Size([64, 4216]) (seq=64) [INFO] [OmniXPU] attention fallback: mask=torch.Size([8, 8]) (seq=8) 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:04<00:00, 1.76it/s] VRAM清理完成 [卸载模型: True, 清空缓存: True] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=1788.6MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 644MB Staged. 0 patches attached. Force pre-loaded 68 weights: 102 KB. [INFO] VRAMdebug: free memory before: 33,529,954,816 [INFO] VRAMdebug: free memory after: 33,530,075,648 [INFO] VRAMdebug: freed memory: 120,832 [INFO] Prompt executed in 99.59 seconds [INFO] Using RAM pressure cache. [INFO] got prompt [INFO] Model storage policy: fast_disk=False paths=['<ComfyUI>/models/vae/qwen_image_2.1_vae_bf16.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 [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_8b_int8_convrot.safetensors'] [INFO] CLIP/text encoder model load device: xpu:0, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=1214.1MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 644MB Staged. 0 patches attached. Force pre-loaded 68 weights: 102 KB. [INFO] Requested to load QwenImage21TEModel_ [INFO] Model QwenImage21TEModel_ prepared for dynamic VRAM loading. 8916MB Staged. 0 patches attached. Force pre-loaded 255 weights: 849 KB. [INFO] Model QwenImage21TEModel_ prepared for dynamic VRAM loading. 8916MB Staged. 0 patches attached. Force pre-loaded 255 weights: 849 KB. [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: nvfp4, float8_e4m3fn, float8_e5m2, mxfp8, w6a8_int8, asym_w4a8_int8, 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>/qwen_image_2.1_int8_convrot.safetensors'] [INFO] [OmniXPU] LoRA memory budget cached: source=152.0MiB/384 tensors model_added=152.0MiB model_budget=152.0MiB clip_added=0.0MiB clip_budget=0.0MiB [INFO] [OmniXPU] LoRA model budget applied: cached=152.0MiB/384 tensors requested=1971.4->2123.4MiB minimum=985.7->1137.7MiB xpu_memory=available xpu_allocated=1.6MiB xpu_reserved=4.0MiB xpu_free=19567.8MiB xpu_total=31977.4MiB [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=1137.7MiB [INFO] Requested to load QwenImage21 [INFO] Model QwenImage21 prepared for dynamic VRAM loading. 6920MB Staged. 192 patches attached. Force pre-loaded 65 weights: 63 KB. [INFO] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=1.7MiB xpu_reserved=4.0MiB xpu_free=19568.5MiB xpu_total=31977.4MiB 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:04<00:00, 1.93it/s] VRAM清理完成 [卸载模型: True, 清空缓存: True] [INFO] [OmniXPU] AIMDO inference budget: devices=0 minimum=1788.6MiB [INFO] Requested to load WanVAE [INFO] Model WanVAE prepared for dynamic VRAM loading. 644MB Staged. 0 patches attached. Force pre-loaded 68 weights: 102 KB. [INFO] VRAMdebug: free memory before: 33,529,954,816 [INFO] VRAMdebug: free memory after: 33,530,075,648 [INFO] VRAMdebug: freed memory: 120,832 [INFO] Prompt executed in 20.88 seconds [INFO] Using RAM pressure cache.