From d7f80d4a41e00c6c8528a08486c66f647861583b Mon Sep 17 00:00:00 2001 From: bymyself Date: Sun, 22 Jun 2025 17:04:48 -0700 Subject: [PATCH] [docs] add ControlNet architecture mismatch error to troubleshooting Adds the common 'mat1 and mat2 shapes cannot be multiplied' error message that occurs when mixing SD1.5 ControlNets with SDXL checkpoints. Updates both English and Chinese documentation. Fixes #8463 --- troubleshooting/model-issues.mdx | 7 ++++++- zh-CN/troubleshooting/model-issues.mdx | 9 ++++++++- 2 files changed, 14 insertions(+), 2 deletions(-) diff --git a/troubleshooting/model-issues.mdx b/troubleshooting/model-issues.mdx index 7a5be5a11..430adadb2 100644 --- a/troubleshooting/model-issues.mdx +++ b/troubleshooting/model-issues.mdx @@ -15,6 +15,7 @@ sidebarTitle: "Model Issues" - `Given groups=1, weight of size [320, 4, 3, 3], expected input[2, 16, 192, 128] to have 4 channels, but got 16 channels instead` - `The size of tensor a (49) must match the size of tensor b (16) at non-singleton dimension 1` - `Tensors must have same number of dimensions: got 2 and 3` +- `mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320)` **Root cause:** Using models from different architecture families together @@ -25,6 +26,7 @@ sidebarTitle: "Model Issues" - **SD1.5 models** use 4-channel latent space with single CLIP ViT-L/14 text encoder - **SDXL models** use 4-channel latent space with dual text encoders (CLIP ViT-L/14 + OpenCLIP ViT-bigG/14) - **SD3 models** use 16-channel latent space with triple text encoder conditioning (CLIP-L + OpenCLIP bigG + T5-XXL) + - **ControlNet models** must match the architecture of the base checkpoint (SD1.5 ControlNets only work with SD1.5 checkpoints, SDXL ControlNets only work with SDXL checkpoints, etc.) 2. **Common mismatch scenarios and fixes:** @@ -42,8 +44,11 @@ sidebarTitle: "Model Issues" **ControlNet architecture mismatch:** ``` - Problem: SD1.5 ControlNet with different architecture checkpoint + Problem: SD1.5 ControlNet with SDXL checkpoint (or vice versa) + Error: "mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320)" Fix: Use ControlNet models designed for your checkpoint architecture + - SD1.5 checkpoints require SD1.5 ControlNets + - SDXL checkpoints require SDXL ControlNets ``` 3. **Quick diagnostics:** diff --git a/zh-CN/troubleshooting/model-issues.mdx b/zh-CN/troubleshooting/model-issues.mdx index 20ab42c74..4e066d979 100644 --- a/zh-CN/troubleshooting/model-issues.mdx +++ b/zh-CN/troubleshooting/model-issues.mdx @@ -13,6 +13,9 @@ sidebarTitle: "模型问题" - `Given groups=1, weight of size [64, 4, 3, 3], expected input[1, 16, 128, 128] to have 4 channels, but got 16 channels instead` - `Given groups=1, weight of size [4, 4, 1, 1], expected input[1, 16, 144, 112] to have 4 channels, but got 16 channels instead` - `Given groups=1, weight of size [320, 4, 3, 3], expected input[2, 16, 192, 128] to have 4 channels, but got 16 channels instead` +- `The size of tensor a (49) must match the size of tensor b (16) at non-singleton dimension 1` +- `Tensors must have same number of dimensions: got 2 and 3` +- `mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320)` **根本原因：** 将来自不同架构系列的模型混合使用 @@ -23,6 +26,7 @@ sidebarTitle: "模型问题" - **SD1.5 模型**使用 4 通道潜在空间，配合单个 CLIP ViT-L/14 文本编码器 - **SDXL 模型**使用 4 通道潜在空间，配合双文本编码器（CLIP ViT-L/14 + OpenCLIP ViT-bigG/14） - **SD3 模型**使用 16 通道潜在空间，配合三重文本编码器调节（CLIP-L + OpenCLIP bigG + T5-XXL） + - **ControlNet 模型**必须与基础检查点的架构匹配（SD1.5 ControlNet 仅适用于 SD1.5 检查点，SDXL ControlNet 仅适用于 SDXL 检查点，等等） 2. **常见不匹配场景和修复：** @@ -40,8 +44,11 @@ sidebarTitle: "模型问题" **ControlNet 架构不匹配：** ``` - 问题：SD1.5 ControlNet 与不同架构检查点 + 问题：SD1.5 ControlNet 与 SDXL 检查点（或反之） + 错误："mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320)" 修复：使用为您的检查点架构设计的 ControlNet 模型 + - SD1.5 检查点需要 SD1.5 ControlNet + - SDXL 检查点需要 SDXL ControlNet ``` 3. **快速诊断：**