202611031216_Moss文生图8步+TE2b优化
2026-11-03 12:16:05 · sess-20261103-105303 · 工作流 Moss
success重载omnixpu加速B58016G
总耗时
111.89s
采样耗时
32s
采样步数
8 / 8
每步采样
4.07s/it
模型载入
8.10GB
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 | 41.243 | 125 | policy 相邻 |
| 文本编码器 text_encoder | qwen3.5-2b_int8_linear.safetensors | #3 MossCLIPLoader | — | — | — | — |
| 文本编码器 text_encoder | qwen3.5-2b-int8.safetensors ext/qwen3.5-2b-int8.safetensors | #3 MossCLIPLoader Qwen35TEModel_ | 2973.7 | 0.080 | 37172 | policy 相邻 |
| VAE vae | flux2-vae.safetensors | #4 VAELoader AutoencoderKL | 160.3 | 0.381 | 421 | 未识别 |
| LoRA lora | MOSS-Image-SFT-8-steps.safetensors Moss/MOSS-Image-SFT-8-steps.safetensors | #2 LoraLoaderModelOnly | — | — | — | — |
共 5 行(图中每一处模型引用各占一行,不合并):3 行在执行日志里有载入数据,2 行只有结构引用。
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导出
| 画布节点数(UI 图) | 13 |
|---|---|
| 实际执行节点数(API 图) | 13 |
| 节点状态分布(UI 图 mode) | active 13 · muted 0 · bypassed 0 |
| 被 bypass 的节点 | 无 |
| 被 mute 的节点 | 无 |
节点详情 13 个节点 · 4 个插件
| 节点名 | 插件名 | 插件版本 |
|---|---|---|
CLIPTextEncode 负面提示词 | comfy-core | 0.41.0 |
CLIPTextEncode 提示词 | comfy-core | 0.41.0 |
EmptyLatentImage 空Latent图像 | comfy-core | 0.41.0 |
KSampler K采样器 | comfy-core | 0.41.0 |
LoraLoaderModelOnly LoRA加载器(仅模型) | comfy-core | 0.41.0 |
ResolutionSelector 分辨率选择器 | comfy-core | 0.41.0 |
SaveImage 保存图像 | comfy-core | 0.41.0 |
UNETLoader UNet加载器 | comfy-core | 0.41.0 |
VAEDecode VAE解码 | comfy-core | 0.41.0 |
VAELoader 加载VAE | comfy-core | 0.41.0 |
LogCollection | ComfyUI-FullLog | — |
MossCLIPLoader Load CLIP (MOSS + projection) | moss-image-mini-encoder 1cyberlangke1 | ee1855e5a699… |
Seed (rgthree) | rgthree-comfy | 8ff50e452188… |
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+4) |
| 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 | 0.43.1 preview=Latent2RGB |
| 插件数 | 67 |
| 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 15 个 adapter applied、4 个 skipped |
| aimdo | comfy-aimdo 0.6.0 未启用 (disabled by env) |
| kitchen XPU | comfy-kitchen 0.2.40 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 17 项
| 变量名 | 值 |
|---|---|
OCL_ICD_FILENAMES | ;F:\Intel-oneAPI\compiler\latest\bin\intelocl64.dll |
OMNIXPU_DEBUG | 0 |
OMNIXPU_ENABLE | 1 |
ONEAPI_CLANG_VERSION | 22 |
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 |
|---|---|---|---|
| 12:16:06.755 | 0.74 | 0.74 | 下发存储策略 · qwen3.5-2b-int8.safetensors |
| 12:16:20.129 | 14.12 | 13.38 | 请求加载 Qwen35TEModel_ |
| 12:16:20.209 | 14.20 | 0.08 | qwen3.5-2b-int8.safetensors 就绪(2973.72 MB) |
| 12:16:36.371 | 30.36 | 16.16 | 下发存储策略 · moss_image_sft_int8_convrot.safetensors |
| 12:16:36.560 | 30.55 | 0.19 | 请求加载 MOSSImage |
| 12:17:17.803 | 71.79 | 41.24 | moss_image_sft_int8_convrot.safetensors 就绪(5162.83 MB) |
| 12:17:51.531 | 105.52 | 33.73 | 请求加载 AutoencoderKL |
| 12:17:51.912 | 105.90 | 0.38 | AutoencoderKL 就绪(160.31 MB) |
| 12:17:57.682 | 111.67 | 5.77 | 执行完成 |
原始日志内嵌,不依赖 user/comfyui.log
来源
comfyui.log · 24 行 · 日志自报耗时 111.89 s原始日志全文
[2026-11-03 12:16:06.012] Found quantization metadata version 1[2026-11-03 12:16:06.013] Using MixedPrecisionOps for text encoder[2026-11-03 12:16:06.755] Model storage policy: fast_disk=False paths=['E:\\model\\qwen3.5-2b-int8.safetensors'][2026-11-03 12:16:20.129] Requested to load Qwen35TEModel_[2026-11-03 12:16:20.209] loaded completely; 2973.72 MB loaded, full load: True[2026-11-03 12:16:20.226] CLIP/text encoder model load device: cpu, offload device: cpu, current: cpu, dtype: torch.float16[2026-11-03 12:16:36.353] Found quantization metadata version 1[2026-11-03 12:16:36.353] Detected mixed precision quantization[2026-11-03 12:16:36.354] Using mixed precision operations[2026-11-03 12:16:36.354] Native ops: int8_tensorwise , emulated ops: mxfp8, w6a8_int8, float8_e5m2, nvfp4, float8_e4m3fn, convrot_w4a4, asym_w4a8_int8[2026-11-03 12:16:36.368] model weight dtype torch.bfloat16, manual cast: torch.bfloat16[2026-11-03 12:16:36.370] model_type FLOW[2026-11-03 12:16:36.371] Model storage policy: fast_disk=False paths=['E:\\model\\moss_image_sft_int8_convrot.safetensors'][2026-11-03 12:16:36.506] [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[2026-11-03 12:16:36.559] [OmniXPU] LoRA model budget applied: cached=1134.0MiB/504 tensors requested=8656.6->9790.6MiB minimum=4328.3->5462.3MiB xpu_memory=available xpu_allocated=166.9MiB xpu_reserved=172.0MiB xpu_free=8776.5MiB xpu_total=11946.2MiB[2026-11-03 12:16:36.560] Requested to load MOSSImage[2026-11-03 12:17:17.803] loaded completely; 5459.38 MB usable, 5162.83 MB loaded, full load: True[2026-11-03 12:17:17.887] [OmniXPU] LoRA model load complete: xpu_memory=available xpu_allocated=5163.3MiB xpu_reserved=5366.0MiB xpu_free=3686.5MiB xpu_total=11946.2MiB[2026-11-03 12:17:50.618] 100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:32<00:00, 3.72s/it]100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:32<00:00, 4.07s/it][2026-11-03 12:17:51.531] Requested to load AutoencoderKL[2026-11-03 12:17:51.912] loaded completely; 5667.06 MB usable, 160.31 MB loaded, full load: True[2026-11-03 12:17:57.682] Prompt executed in 111.89 seconds


