202611180044_MossEdit动漫转真人+TE4b优化_omnixpu加速_B580_16G_重载
2026-11-18 00:44:49 · sess-20261117-232156 · 工作流 image_moss_image_edit
success重载omnixpu加速B58016G
总耗时
232.71s
采样耗时
10s
采样步数
8 / 8
每步采样
1.27s/it
模型载入
9.10GB
单次峰值
5.04GB
生成物

模型加载模型文件 / 加载节点来自工作流结构(API 图);载入 MB / 耗时 / MB/s / 路径来源来自执行日志
| 角色 | 模型文件 | 加载节点 | 载入 MB | 耗时 s | MB/s | 路径来源 |
|---|---|---|---|---|---|---|
| 底模 base | moss_image_sft_int8_convrot.safetensors ext/moss_image_sft_int8_convrot.safetensors | #1 UNETLoader MOSSImage | 5162.8 | 61.229 | 84 | policy 相邻 |
| 文本编码器 text_encoder | qwen3.5-4b_w4a8_linear.safetensors | #7 MossCLIPLoader | — | — | — | — |
| 文本编码器 text_encoder | qwen3.5-4b-w4a8.safetensors ext/qwen3.5-4b-w4a8.safetensors | #7 MossCLIPLoader Qwen35TEModel_ | 3839.7 | 0.164 | 23412 | policy 相邻 |
| VAE vae | flux2-vae.safetensors | #5 VAELoader AutoencoderKL | 160.3 | 0.107 | 1498 | 按角色匹配 |
| VAE vae | (工作流执行日志未记录文件名) | 图中无对应加载节点(间接加载) AutoencoderKL | 160.3 | 0.088 | 1822 | 未识别 |
| LoRA lora | Moss_SFT_Edit_A2R_v1_000006000.safetensors Moss/Moss_SFT_Edit_A2R_v1_000006000.safetensors | #2 LoraLoaderModelOnly | — | — | — | 工作流结构 |
| LoRA lora | MOSS-Image-SFT-8-steps.safetensors Moss/MOSS-Image-SFT-8-steps.safetensors | #3 LoraLoaderModelOnly | — | — | — | 工作流结构 |
| LoRA lora | moss_image_edit_lora_rank_64_bf16.safetensors Moss/moss_image_edit_lora_rank_64_bf16.safetensors | #4 LoraLoaderModelOnly | — | — | — | 工作流结构 |
共 8 行:底模 1 · 文本编码器 2 · VAE 2 · LoRA 3。模型文件 / 加载节点来自工作流结构(API 图)。载入 MB / 耗时 / MB/s 来自工作流执行日志(`Requested to load` → `loaded completely`);同一文件一次执行里加载多次时合并为一行、数值是累计值,相加仍等于 KPI 的「模型载入」。LoRA 是给底模打补丁、不走模型加载,执行日志里没有它的载入数据,所以这三列留空——它的文件名只能来自结构。
MB/s 用于区分「内存回迁」(GB/s 级)与「磁盘重读」(数百 MB/s)。text_encoder 在某些栈上 device_to=cpu,它的加载耗时接近 0,不代表省了显存搬运。
MB/s 用于区分「内存回迁」(GB/s 级)与「磁盘重读」(数百 MB/s)。text_encoder 在某些栈上 device_to=cpu,它的加载耗时接近 0,不代表省了显存搬运。
工作流结构UI 图含 bypass 节点,API 图是真正执行的子集工作流导出
| 画布节点数(UI 图) | 17 |
|---|---|
| 实际执行节点数(API 图) | 17 |
| 节点状态分布(UI 图 mode) | active 17 · muted 0 · bypassed 0 |
| 被 bypass 的节点 | 无 |
| 被 mute 的节点 | 无 |
节点详情 17 个节点 · 6 个插件
| 节点名 | 插件名 | 插件版本 |
|---|---|---|
EmptyFlux2LatentImage 空Latent图像(Flux2) | comfy-core | 0.43.1 |
KSampler K采样器 | comfy-core | 0.43.1 |
LoadImage 加载图像 | comfy-core | 0.27.1 |
LoraLoaderModelOnly LoRA加载器(仅模型) | comfy-core | 0.43.1 |
LoraLoaderModelOnly LoRA加载器(仅模型) | comfy-core | 0.41.0 |
LoraLoaderModelOnly LoRA加载器(仅模型) | comfy-core | 0.43.1 |
SaveImage 保存图像 | comfy-core | 0.43.1 |
TextEncodeMOSSImageEdit 文本编码 MOSS-Image 编辑 | comfy-core | 0.44.0 |
TextEncodeMOSSImageEdit 文本编码 MOSS-Image 编辑 | comfy-core | 0.44.0 |
UNETLoader UNet加载器 | comfy-core | 0.43.1 |
VAEDecodeTiled VAE解码(分块) | comfy-core | 0.11.1 |
VAELoader 加载VAE | comfy-core | 0.43.1 |
LogCollection | ComfyUI-FullLog | — |
VRAM_Debug | comfyui-kjnodes | 9b565354e32a… |
LayerUtility: ImageScaleByAspectRatio V2 图层工具:按宽高比缩放 V2 | comfyui_layerstyle | d94bef1ee5ed… |
VRAMCleanup | comfyui_memory_cleanup | 1.1.3 |
MossCLIPLoader Load CLIP (MOSS + projection) | moss-image-mini-encoder 1cyberlangke1 | ee1855e5a699… |
运行环境
| 系统 | Windows-11-10.0.22000-SP0 · AMD64 · build 10.0.22000 |
|---|---|
| 物理内存 | 15.9 GB (16G) |
| GPU | Intel(R) Arc(TM) B580 Graphics (0xE20B, 11946.2 MB, driver 1.17.39894+6) |
| 平台 | Intel(R) oneAPI Unified Runtime over Level-Zero V2 |
| 计算单元 | 160 |
| Python | 3.13.11 |
| torch | 2.15.0+xpu (xpu available: True, devices: 1) |
| ComfyUI | 0.45.2 preview=Latent2RGB |
| 插件数 | 69 |
| 加速栈 | |
| 加速方案 | omnixpu加速 (工作流执行日志中 omni_xpu_kernel 0.2.0b3+torch215.bmg 生效,且出现 [OmniXPU] 插件标记) |
| OmniXPU 内核 | omni_xpu_kernel 0.2.0b3+torch215.bmg 已生效:sdp, norm, rotary, linear_fp8, int8, layout |
| OmniXPU 插件 | ComfyUI-OmniXPU 16 个 adapter applied、4 个 skipped |
| aimdo | comfy-aimdo 0.6.0 未启用 (disabled by env) |
| kitchen XPU | comfy-kitchen 0.2.41 xpu 后端可用 (adaln, apply_rope, apply_rope1, apply_rope1_… 共 50 项) |
加速方案 = 内核生效行 + 插件
[OmniXPU] 标记,两处都在工作流执行日志里才判「omnixpu加速」;aimdo / kitchen 是可选栈,不影响方案判定,也不进文件名。运行时环境变量 18 项
| 变量名 | 值 |
|---|---|
OCL_ICD_FILENAMES | ;F:\Intel-oneAPI\compiler\latest\bin\intelocl64.dll |
OMNIXPU_DEBUG | 0 |
OMNIXPU_ENABLE | 1 |
OMNIXPU_QWEN_IMAGE21_QKV | 1 |
ONEAPI_CLANG_VERSION | 23 |
ONEAPI_DEVICE_SELECTOR | level_zero:0 |
ONEAPI_PATH | F:\Intel-oneAPI |
ONEAPI_ROOT | F:\Intel-oneAPI |
SYCL_CACHE_MAX_SIZE | 8192 |
SYCL_CACHE_THRESHOLD | 7 |
SYCL_DEVICE_FILTER | level_zero |
SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS | 0 |
SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY | 1 |
SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS | 1 |
UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS | 1 |
UR_L0_USE_IMMEDIATE_COMMANDLISTS | 1 |
ZE_DEVICE_SLEEP | 0 |
ZE_ENABLE_PCI_ID_DEVICE_ORDER | 1 |
时间轴
| 时刻 | 相对起点 s | 增量 s | 事件 |
|---|---|---|---|
| 00:44:49.444 | 0.00 | 0.00 | 下发存储策略 · flux2-vae.safetensors |
| 00:44:51.756 | 2.31 | 2.31 | 下发存储策略 · qwen3.5-4b-w4a8.safetensors |
| 00:45:20.235 | 30.79 | 28.48 | 请求加载 Qwen35TEModel_ |
| 00:45:20.399 | 30.95 | 0.16 | qwen3.5-4b-w4a8.safetensors 就绪(3839.65 MB) |
| 00:46:21.942 | 92.50 | 61.55 | 请求加载 AutoencoderKL |
| 00:46:22.049 | 92.61 | 0.11 | flux2-vae.safetensors 就绪(160.31 MB) |
| 00:47:10.485 | 141.04 | 48.43 | 下发存储策略 · moss_image_sft_int8_convrot.safetensors |
| 00:47:11.284 | 141.84 | 0.80 | 请求加载 MOSSImage |
| 00:48:12.513 | 203.07 | 61.23 | MOSSImage 就绪(5162.83 MB) |
| 00:48:38.623 | 229.18 | 26.11 | 请求加载 AutoencoderKL |
| 00:48:38.711 | 229.27 | 0.09 | flux2-vae.safetensors 就绪(160.31 MB) |
| 00:48:42.050 | 232.61 | 3.34 | 执行完成 |
原始日志内嵌,不依赖 user/comfyui.log
来源
user/comfyui.log · 37 行 · 日志自报耗时 232.71 s原始日志全文
[2026-11-18 00:44:49.444] Model storage policy: fast_disk=False paths=['F:\\ComfyUI-aki-v3\\ComfyUI\\models\\vae\\flux2-vae.safetensors'][2026-11-18 00:44:49.450] Using pytorch attention in VAE[2026-11-18 00:44:49.455] Using pytorch attention in VAE[2026-11-18 00:44:50.014] VAE load device: xpu:0, offload device: cpu, dtype: torch.float16[2026-11-18 00:44:50.414] # 😺dzNodes: LayerStyle -> ImageScaleByAspectRatio V2 Processed 1 image(s).[2026-11-18 00:44:50.892] Found quantization metadata version 1[2026-11-18 00:44:50.894] Using MixedPrecisionOps for text encoder[2026-11-18 00:44:51.756] Model storage policy: fast_disk=False paths=['E:\\model\\qwen3.5-4b-w4a8.safetensors'][2026-11-18 00:45:20.235] Requested to load Qwen35TEModel_[2026-11-18 00:45:20.399] loaded completely; 3839.65 MB loaded, full load: True[2026-11-18 00:45:20.525] CLIP/text encoder model load device: cpu, offload device: cpu, current: cpu, dtype: torch.float16[2026-11-18 00:46:21.942] Requested to load AutoencoderKL[2026-11-18 00:46:22.049] loaded completely; 5719.06 MB usable, 160.31 MB loaded, full load: True[2026-11-18 00:47:10.466] Found quantization metadata version 1[2026-11-18 00:47:10.466] Detected mixed precision quantization[2026-11-18 00:47:10.466] Using mixed precision operations[2026-11-18 00:47:10.467] Native ops: int8_tensorwise, asym_w4a8_int8 , emulated ops: mxfp8, nvfp4, convrot_w4a4, w6a8_int8, float8_e4m3fn, float8_e5m2[2026-11-18 00:47:10.482] model weight dtype torch.bfloat16, manual cast: torch.bfloat16[2026-11-18 00:47:10.485] model_type FLOW[2026-11-18 00:47:10.485] Model storage policy: fast_disk=False paths=['E:\\model\\moss_image_sft_int8_convrot.safetensors'][2026-11-18 00:47:10.787] [OmniXPU] LoRA memory budget cached: source=283.5MiB/504 tensors model_added=283.5MiB model_budget=283.5MiB clip_added=0.0MiB clip_budget=0.0MiB[2026-11-18 00:47:11.015] [OmniXPU] LoRA memory budget cached: source=276.8MiB/492 tensors model_added=276.8MiB model_budget=560.2MiB clip_added=0.0MiB clip_budget=0.0MiB[2026-11-18 00:47:11.256] [OmniXPU] LoRA memory budget cached: source=1134.2MiB/756 tensors model_added=1134.0MiB model_budget=1694.2MiB clip_added=0.0MiB clip_budget=0.0MiB[2026-11-18 00:47:11.283] [OmniXPU] LoRA model budget applied: cached=1694.2MiB/1500 tensors requested=7043.0->8737.3MiB minimum=3521.5->5215.8MiB xpu_memory=available xpu_allocated=166.9MiB xpu_reserved=1716.0MiB xpu_free=7150.5MiB xpu_total=11946.2MiB[2026-11-18 00:47:11.284] Requested to load MOSSImage[2026-11-18 00:48:12.513] loaded completely; 5705.93 MB usable, 5162.83 MB loaded, full load: True[2026-11-18 00:48:12.532] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=5163.3MiB xpu_reserved=5366.0MiB xpu_free=3587.8MiB xpu_total=11946.2MiB[2026-11-18 00:48:29.665] 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:10<00:00, 1.25s/it]100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:10<00:00, 1.27s/it][2026-11-18 00:48:38.611] VRAM清理完成 [卸载模型: True, 清空缓存: True][2026-11-18 00:48:38.623] Requested to load AutoencoderKL[2026-11-18 00:48:38.711] loaded completely; 10102.50 MB usable, 160.31 MB loaded, full load: True[2026-11-18 00:48:41.340] VRAMdebug: free memory before: 12,351,444,992[2026-11-18 00:48:41.820] VRAMdebug: free memory after: 12,525,972,480[2026-11-18 00:48:41.821] VRAMdebug: freed memory: 174,527,488[2026-11-18 00:48:42.050] Prompt executed in 232.71 seconds