今天一位群友在运行AI绘画软件“Stable Diffusion”的时候，结果报错类似如下：
RuntimeError: CUDA error: device kernel image is invalid
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
中文翻译
运行时错误：CUDA错误：设备内核映像无效
CUDA内核错误可能会在其他一些API调用中异步报告，因此下面的 stacktrace 可能不正确。
对于调试，请考虑传递 CUDA_LAUNCH_BLOCKING=1 。
使用“TORCH_USE_CUDA_DSA”进行编译以启用设备端断言。
原因
通过上面的报错，也能知道一个最主要的原因就是：CUDA错误：设备内核映像无效。
后来，我又向这位朋友要到了整个“控制台”的代码，如下：
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Commit hash: 22bcc7be428c94e9408f589966c2040187245d81
Installing requirements for Web UI
Launching Web UI with arguments: --medvram --theme dark --xformers --api --autolaunch
AUTOMATIC1111/stable-diffusion-webui packed by bilibili@秋葉aaaki
本整合包完全免费，严禁倒卖。若您付费获得本软件请立刻举报商家。
[AddNet] Updating model hashes...
[AddNet] Updating model hashes...
Loading weights [7f96a1a9ca] from G:\SD\sd-webui-aki-v4\models\Stable-diffusion\anything-v5-PrtRE.safetensors
Creating model from config: G:\SD\sd-webui-aki-v4\configs\v1-inference.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Loading VAE weights specified in settings: G:\SD\sd-webui-aki-v4\models\VAE\animevae.pt
Applying xformers cross attention optimization.
Textual inversion embeddings loaded(1): EasyNegative
Model loaded in 71.2s (load weights from disk: 1.6s, create model: 1.1s, apply weights to model: 55.3s, apply half(): 5.6s, load VAE: 6.8s, hijack: 0.1s, load textual inversion embeddings: 0.5s).
Running on local URL: http://127.0.0.1:7860
To create a public link, set `share=True` in `launch()`.
Startup time: 98.8s (import torch: 4.3s, import gradio: 3.9s, import ldm: 1.8s, other imports: 4.8s, setup codeformer: 0.3s, load scripts: 3.1s, load SD checkpoint: 71.3s, create ui: 8.2s, gradio launch: 0.8s, scripts app_started_callback: 0.1s).
Error completing request
Arguments: ('task(q4qks2qdumkiupn)', '', '', [], 20, 0, False, False, 1, 1, 7, -1.0, -1.0, 0, 0, 0, False, 512, 512, False, 0.7, 2, 'Latent', 0, 0, 0, [], 0, False, 'MultiDiffusion', False, 10, 1, 1, 64, False, True, 1024, 1024, 96, 96, 48, 1, 'None', 2, False, False, False, False, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, False, True, True, False, 1536, 96, False, False, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, None, 'Refresh models', <scripts.external_code.ControlNetUnit object at 0x0000012F44116620>, <scripts.external_code.ControlNetUnit object at 0x0000012F44116650>, <scripts.external_code.ControlNetUnit object at 0x0000012F441166E0>, <scripts.external_code.ControlNetUnit object at 0x0000012F44116770>, False, False, 'positive', 'comma', 0, False, False, '', 1, '', 0, '', 0, '', True, False, False, False, 0, None, False, None, False, None, False, None, False, 50) {}
Traceback (most recent call last):
File "G:\SD\sd-webui-aki-v4\modules\call_queue.py", line 56, in f
res = list(func(*args, **kwargs))
File "G:\SD\sd-webui-aki-v4\modules\call_queue.py", line 37, in f
res = func(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\modules\txt2img.py", line 56, in txt2img
processed = process_images(p)
File "G:\SD\sd-webui-aki-v4\modules\processing.py", line 503, in process_images
res = process_images_inner(p)
File "G:\SD\sd-webui-aki-v4\modules\processing.py", line 642, in process_images_inner
uc = get_conds_with_caching(prompt_parser.get_learned_conditioning, negative_prompts, p.steps, cached_uc)
File "G:\SD\sd-webui-aki-v4\modules\processing.py", line 587, in get_conds_with_caching
cache[1] = function(shared.sd_model, required_prompts, steps)
File "G:\SD\sd-webui-aki-v4\modules\prompt_parser.py", line 140, in get_learned_conditioning
conds = model.get_learned_conditioning(texts)
File "G:\SD\sd-webui-aki-v4\repositories\stable-diffusion-stability-ai\ldm\models\diffusion\ddpm.py", line 669, in get_learned_conditioning
c = self.cond_stage_model(c)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack_clip.py", line 229, in forward
z = self.process_tokens(tokens, multipliers)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack_clip.py", line 254, in process_tokens
z = self.encode_with_transformers(tokens)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack_clip.py", line 302, in encode_with_transformers
outputs = self.wrapped.transformer(input_ids=tokens, output_hidden_states=-opts.CLIP_stop_at_last_layers)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1538, in _call_impl
result = forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\transformers\models\clip\modeling_clip.py", line 811, in forward
return self.text_model(
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\transformers\models\clip\modeling_clip.py", line 708, in forward
hidden_states = self.embeddings(input_ids=input_ids, position_ids=position_ids)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\transformers\models\clip\modeling_clip.py", line 223, in forward
inputs_embeds = self.token_embedding(input_ids)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack.py", line 234, in forward
inputs_embeds = self.wrapped(input_ids)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\sparse.py", line 162, in forward
return F.embedding(
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\functional.py", line 2210, in embedding
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: CUDA error: device kernel image is invalid
提示：Python 运行时抛出了一个异常。请检查疑难解答页面。
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
Error completing request
Arguments: ('task(12np73tygyxboil)', '猪', '', [], 20, 0, False, False, 1, 1, 7, -1.0, -1.0, 0, 0, 0, False, 512, 512, False, 0.7, 2, 'Latent', 0, 0, 0, [], 0, False, 'MultiDiffusion', False, 10, 1, 1, 64, False, True, 1024, 1024, 96, 96, 48, 1, 'None', 2, False, False, False, False, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, 0.4, 0.4, 0.2, 0.2, '', '', 'Background', 0.2, -1.0, False, False, True, True, False, 1536, 96, False, False, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, 'LoRA', 'None', 1, 1, None, 'Refresh models', <scripts.external_code.ControlNetUnit object at 0x0000012F44116620>, <scripts.external_code.ControlNetUnit object at 0x0000012F44116650>, <scripts.external_code.ControlNetUnit object at 0x0000012F441166E0>, <scripts.external_code.ControlNetUnit object at 0x0000012F44116770>, False, False, 'positive', 'comma', 0, False, False, '', 1, '', 0, '', 0, '', True, False, False, False, 0, None, False, None, False, None, False, None, False, 50) {}
Traceback (most recent call last):
File "G:\SD\sd-webui-aki-v4\modules\call_queue.py", line 56, in f
res = list(func(*args, **kwargs))
File "G:\SD\sd-webui-aki-v4\modules\call_queue.py", line 37, in f
res = func(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\modules\txt2img.py", line 56, in txt2img
processed = process_images(p)
File "G:\SD\sd-webui-aki-v4\modules\processing.py", line 503, in process_images
res = process_images_inner(p)
File "G:\SD\sd-webui-aki-v4\modules\processing.py", line 642, in process_images_inner
uc = get_conds_with_caching(prompt_parser.get_learned_conditioning, negative_prompts, p.steps, cached_uc)
File "G:\SD\sd-webui-aki-v4\modules\processing.py", line 587, in get_conds_with_caching
cache[1] = function(shared.sd_model, required_prompts, steps)
File "G:\SD\sd-webui-aki-v4\modules\prompt_parser.py", line 140, in get_learned_conditioning
conds = model.get_learned_conditioning(texts)
File "G:\SD\sd-webui-aki-v4\repositories\stable-diffusion-stability-ai\ldm\models\diffusion\ddpm.py", line 669, in get_learned_conditioning
c = self.cond_stage_model(c)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack_clip.py", line 229, in forward
z = self.process_tokens(tokens, multipliers)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack_clip.py", line 254, in process_tokens
z = self.encode_with_transformers(tokens)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack_clip.py", line 302, in encode_with_transformers
outputs = self.wrapped.transformer(input_ids=tokens, output_hidden_states=-opts.CLIP_stop_at_last_layers)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1538, in _call_impl
result = forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\transformers\models\clip\modeling_clip.py", line 811, in forward
return self.text_model(
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\transformers\models\clip\modeling_clip.py", line 708, in forward
hidden_states = self.embeddings(input_ids=input_ids, position_ids=position_ids)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\transformers\models\clip\modeling_clip.py", line 223, in forward
inputs_embeds = self.token_embedding(input_ids)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\modules\sd_hijack.py", line 234, in forward
inputs_embeds = self.wrapped(input_ids)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\modules\sparse.py", line 162, in forward
return F.embedding(
File "G:\SD\sd-webui-aki-v4\py310\lib\site-packages\torch\nn\functional.py", line 2210, in embedding
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: CUDA error: device kernel image is invalid
提示：Python 运行时抛出了一个异常。请检查疑难解答页面。
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
解决方法
既然说是“CUDA错误：设备内核映像无效。”，那我们就去手工安装CUDA。
具体的操作步骤如下：
1、命令行运行命令：nvidia-smi，查看自己显卡支持的 CUDA版本，升级显卡驱动有可能会让你支持更高版本的 CUDA。
2、然后前往英伟达 CUDA 官网，下载对应版本。
英伟达 CUDA 官方下载网址：https://developer.nvidia.com/cuda-downloads
英伟达CUDA各版本官方下载地址：https://developer.nvidia.com/cuda-toolkit-archive
注意：
你对应的版本号最高的版本，比如我的是12.0的，那就下12.0.1（这里最后的.1意思是，11.7版本的1号升级版），总之一句话，下载的版本比12.0低就行了。
3、选你自己的操作系统版本，选择离线安装包“exe [local]”。
（这张图忘记截了，从网上找了一张类似的，区别只是图中版本是11.7.1）
4、后面只需要安装即可！
如果你还有不懂的，具体操作请参考：CUDA下载与安装教程
