kapibala

> 1. 可以。并不是强制的，只是命令行帮你做个预处理。俺都是直接魔改源码、喂 .bin 进去； > 2. 内部的 layout 是 nchw， 命令行习惯应该是 whc。强制的不能瞎写。 但我这样运行ncnn2table /home/lkw/programfiles/hesikeji/tecent/ncnn/build/tools/quantize/ncnn2table 2.7_80x80_MiniFASNetV2-sim-opt.param 2.7_80x80_MiniFASNetV2-sim-opt.bin 8080image.txt 2.7_80x80_MiniFASNetV2.table mean=[] norm=[] shape=[80,80,3] pixel=BGR thread=8 method=kl 会报错：expect 1 means, but got...

@mi804 好兄弟，想问下我采用这种方式： ``` import glob from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig from PIL import Image import torch pipe = QwenImagePipeline.from_pretrained( torch_dtype=torch.bfloat16, device="cuda", model_configs=[ ModelConfig(path=['/home/hadoop-automaterials/dolphinfs_hdd_hadoop-automaterials/likaiwen/PythonProjects/Qwen_image_train/model/model-00001-of-00004.safetensors', '/home/hadoop-automaterials/dolphinfs_hdd_hadoop-automaterials/likaiwen/PythonProjects/Qwen_image_train/model/model-00002-of-00004.safetensors', '/home/hadoop-automaterials/dolphinfs_hdd_hadoop-automaterials/likaiwen/PythonProjects/Qwen_image_train/model/model-00003-of-00004.safetensors', '/home/hadoop-automaterials/dolphinfs_hdd_hadoop-automaterials/likaiwen/PythonProjects/Qwen_image_train/model/model-00004-of-00004.safetensors']), ModelConfig(path= "/home/hadoop-automaterials/dolphinfs_hdd_hadoop-automaterials/yongzhao41/insert-anything/checkpoints/Qwen-Image/vae/diffusion_pytorch_model.safetensors"), ModelConfig(path=['/home/hadoop-automaterials/dolphinfs_hdd_hadoop-automaterials/yongzhao41/insert-anything/checkpoints/Qwen-Image/transformer/diffusion_pytorch_model-00001-of-00009.safetensors', '/home/hadoop-automaterials/dolphinfs_hdd_hadoop-automaterials/yongzhao41/insert-anything/checkpoints/Qwen-Image/transformer/diffusion_pytorch_model-00002-of-00009.safetensors',...