202611020027_Moss文生图_omnixpu加速_B580_16G_冷启+热启.txt · 2026-11-02 00:27
| 阶段 | 耗时 | 步数 | s/it | staged MB | 其他指标 |
|---|---|---|---|---|---|
| 冷启 | 106.81 s | 8 | 1.41 | 5,163† | CUTE ×3 |
| 热启 | 64.03 s | 8 | 1.07 | 5,163† | — |
† 该日志的运行时只打印 loaded completely; … MB loaded,没有 Staged 行;数值为单模型最大驻留量,与 Staged 同一口径,可横向比较。
[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.float16 [INFO] Found quantization metadata version 1 [INFO] Using MixedPrecisionOps for text encoder [INFO] Model storage policy: fast_disk=False paths=['<本地路径>/qwen3.5-2b-int8.safetensors'] [INFO] Requested to load Qwen35TEModel_ [INFO] loaded completely; 2973.72 MB loaded, full load: True [INFO] CLIP/text encoder model load device: cpu, offload device: cpu, current: cpu, dtype: torch.float16 [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: convrot_w4a4, w6a8_int8, asym_w4a8_int8, nvfp4, float8_e4m3fn, float8_e5m2, mxfp8 [INFO] model weight dtype torch.bfloat16, manual cast: torch.bfloat16 [INFO] model_type FLOW [INFO] Model storage policy: fast_disk=False paths=['<本地路径>/moss_image_sft_int8_convrot.safetensors'] [INFO] [OmniXPU] LoRA memory budget cached: source=1134.2MiB/756 tensors model_added=1134.0MiB model_budget=1134.0MiB clip_added=0.0MiB clip_budget=0.0MiB [INFO] [OmniXPU] LoRA model budget applied: cached=1134.0MiB/504 tensors requested=2382.7->3516.7MiB minimum=1191.4->2325.4MiB xpu_memory=available xpu_allocated=0.0MiB xpu_reserved=2.0MiB xpu_free=10253.6MiB xpu_total=11946.2MiB [INFO] Requested to load MOSSImage [INFO] loaded completely; 8596.84 MB usable, 5162.83 MB loaded, full load: True [INFO] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=5163.2MiB xpu_reserved=5366.0MiB xpu_free=4740.9MiB xpu_total=11946.2MiB 0%| | 0/8 [00:00<?, ?it/s][INFO] [OmniXPU] norm first use: RMSNorm shape=torch.Size([1, 281, 3072]) [INFO] [OmniXPU] attention CUTE #1: heads=24 seq=281 dtype=torch.bfloat16 [INFO] [OmniXPU] attention CUTE #2: heads=24 seq=281 dtype=torch.bfloat16 [INFO] [OmniXPU] attention CUTE #3: heads=24 seq=4361 dtype=torch.bfloat16 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:11<00:00, 1.41s/it] [INFO] Requested to load AutoencoderKL [INFO] loaded completely; 6582.71 MB usable, 160.31 MB loaded, full load: True [INFO] Prompt executed in 106.81 seconds [INFO] got prompt [INFO] Found quantization metadata version 1 [INFO] Detected mixed precision quantization [INFO] Using mixed precision operations [INFO] Native ops: int8_tensorwise , emulated ops: convrot_w4a4, w6a8_int8, asym_w4a8_int8, nvfp4, float8_e4m3fn, float8_e5m2, mxfp8 [INFO] model weight dtype torch.bfloat16, manual cast: torch.bfloat16 [INFO] model_type FLOW [INFO] Model storage policy: fast_disk=False paths=['<本地路径>/moss_image_sft_int8_convrot.safetensors'] [INFO] [OmniXPU] LoRA memory budget cached: source=1134.2MiB/756 tensors model_added=1134.0MiB model_budget=1134.0MiB clip_added=0.0MiB clip_budget=0.0MiB [INFO] [OmniXPU] LoRA model budget applied: cached=1134.0MiB/504 tensors requested=2382.7->3516.7MiB minimum=1191.4->2325.4MiB xpu_memory=available xpu_allocated=166.9MiB xpu_reserved=172.0MiB xpu_free=9638.5MiB xpu_total=11946.2MiB [INFO] Requested to load MOSSImage [INFO] loaded completely; 8429.90 MB usable, 5162.83 MB loaded, full load: True [INFO] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=5329.8MiB xpu_reserved=5536.0MiB xpu_free=4395.7MiB xpu_total=11946.2MiB 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:08<00:00, 1.07s/it] [INFO] Prompt executed in 64.03 seconds