910B机器部署comfyui 跑sd_xl_base_1.0.safetensors 推理失败。

一、问题现象（附报错日志上下文）：
910B机器可以部署并运行comfyui,但是推理报错，模型：sd_xl_base_1.0safetensors。（报错指引是算子库问题）。

二、软件版本:
-- CANN 版本: (Ascend-cann-kernels-910b_8.1.RC1.alpha001_linux-aarch64.run, Ascend-cann-toolkit_8.1.RC1.alpha001_linux-aarch64.run) 
--Tensorflow/Pytorch/MindSpore 版本: NO
![输入图片说明](https://foruda.gitee.com/images/1746696786353819790/479f7137_5131031.png "屏幕截图")
--Python 版本 :Python: 3.10.16
-- MindStudio版本 (e.g., MindStudio 2.0.0 (beta3)): NO
--操作系统版本 (e.g., Ubuntu 18.04): Ubuntu 22.04.4 LTS
![输入图片说明](https://foruda.gitee.com/images/1746696860760489154/32a6f459_5131031.png "屏幕截图")
三、测试步骤：
1、安装上述CANN 两个软件包，并测试diffusers 成功。
-- 算子包安装：https://ascend.github.io/docs/sources/ascend/quick_install.html#id2
![输入图片说明](https://foruda.gitee.com/images/1746697130787228937/7f0369ae_5131031.png "屏幕截图")
-- 算子包说明（华为云链接：https://support.huaweicloud.com/bestpractice-modelarts/modelarts_10_01205.html#ZH-CN_TOPIC_0000001967619758__zh-cn_topic_0000001929374254_table16641319162317）：由于没法安装（需要认证才能安装）华为的AscendCloud-6.3.906-xxxx.zip，所以才选择安装上面那个算子库，是否是comfyui 必须要华为云的AIGC算子库呢？
-- 测试diffusers
昇腾开源链接：https://ascend.github.io/docs/sources/Diffusers/install.html
![输入图片说明](https://foruda.gitee.com/images/1746697021410196586/1ddd1876_5131031.png "屏幕截图")

2、git 下载Comfyui，并启动comfyui，测试是可以加载npu的。
-- comfyui 启动信息：
![输入图片说明](https://foruda.gitee.com/images/1746697529800156540/2752501c_5131031.png "屏幕截图")

四、日志信息:
-- comfyui web 页面日志（图片）：
![输入图片说明](https://foruda.gitee.com/images/1746697979403003341/eed11532_5131031.png "屏幕截图")

-- comfyui 后端日志（图片）：

```
got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using pytorch attention in VAE
Using pytorch attention in VAE
VAE load device: npu:0, offload device: cpu, dtype: torch.bfloat16
CLIP/text encoder model load device: npu:0, offload device: cpu, current: cpu, dtype: torch.float16
Requested to load SDXLClipModel
loaded completely 60870.21796875 1560.802734375 True
!!! Exception during processing !!! call aclnnEmbedding failed, detail:EZ9999: Inner Error!
EZ9999: [PID: 1120203] 2025-05-08-17:52:32.176.194 Parse dynamic kernel config fail.
        TraceBack (most recent call last):
       AclOpKernelInit failed opType
       GatherV2AiCore ADD_TO_LAUNCHER_LIST_AICORE failed.

[ERROR] 2025-05-08-17:52:32 (PID:1120203, Device:0, RankID:-1) ERR01100 OPS call acl api failed
Traceback (most recent call last):
  File "/root/zhaoxin/npu-comfyui/ComfyUI/execution.py", line 347, in execute
    output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/execution.py", line 222, in get_output_data
    return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/execution.py", line 194, in _map_node_over_list
    process_inputs(input_dict, i)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/execution.py", line 183, in process_inputs
    results.append(getattr(obj, func)(**inputs))
  File "/root/zhaoxin/npu-comfyui/ComfyUI/nodes.py", line 69, in encode
    return (clip.encode_from_tokens_scheduled(tokens), )
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/sd.py", line 164, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/sd.py", line 226, in encode_from_tokens
    o = self.cond_stage_model.encode_token_weights(tokens)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/sdxl_clip.py", line 59, in encode_token_weights
    g_out, g_pooled = self.clip_g.encode_token_weights(token_weight_pairs_g)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/sd1_clip.py", line 45, in encode_token_weights
    o = self.encode(to_encode)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/sd1_clip.py", line 288, in encode
    return self(tokens)
  File "/root/miniconda3/envs/confyui/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/root/miniconda3/envs/confyui/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/sd1_clip.py", line 250, in forward
    embeds, attention_mask, num_tokens = self.process_tokens(tokens, device)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/sd1_clip.py", line 204, in process_tokens
    tokens_embed = self.transformer.get_input_embeddings()(tokens_embed, out_dtype=torch.float32)
  File "/root/miniconda3/envs/confyui/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/root/miniconda3/envs/confyui/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
    return forward_call(*args, **kwargs)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/ops.py", line 237, in forward
    return self.forward_comfy_cast_weights(*args, **kwargs)
  File "/root/zhaoxin/npu-comfyui/ComfyUI/comfy/ops.py", line 233, in forward_comfy_cast_weights
    return torch.nn.functional.embedding(input, weight, self.padding_idx, self.max_norm, self.norm_type, self.scale_grad_by_freq, self.sparse).to(dtype=output_dtype)
  File "/root/miniconda3/envs/confyui/lib/python3.10/site-packages/torch/nn/functional.py", line 2267, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: call aclnnEmbedding failed, detail:EZ9999: Inner Error!
EZ9999: [PID: 1120203] 2025-05-08-17:52:32.176.194 Parse dynamic kernel config fail.
        TraceBack (most recent call last):
       AclOpKernelInit failed opType
       GatherV2AiCore ADD_TO_LAUNCHER_LIST_AICORE failed.

[ERROR] 2025-05-08-17:52:32 (PID:1120203, Device:0, RankID:-1) ERR01100 OPS call acl api failed

Prompt executed in 10.21 seconds
```

plog日志：

```
[ERROR] OP(1004400,python3):2025-05-08-17:05:30.469.147 [kernel_mgr.cpp:205][NNOP][ParseDynamicKernelConfig][1005106] errno[561108] OpName:[aclnnEmbedding_1_GatherV2AiCore] Parse dynamic kernel config fail.
[ERROR] OP(1004400,python3):2025-05-08-17:05:30.469.326 [kernel_workspace.cpp:91][NNOP][GetWorkspace][1005106] errno[561000] OpName:[aclnnEmbedding_1_GatherV2AiCore] AclOpKernelInit failed opType
[ERROR] OP(1004400,python3):2025-05-08-17:05:30.469.373 [gather_v2.cpp:64][NNOP][GatherV2AiCore][1005106] errno[561103] OpName:[aclnnEmbedding_1_GatherV2AiCore] GatherV2AiCore ADD_TO_LAUNCHER_LIST_AICORE failed.
[ERROR] OP(1004400,python3):2025-05-08-17:05:30.469.401 [aclnn_embedding.cpp:177][NNOP][aclnnEmbeddingGetWorkspaceSize][1005106] errno[561000] OpName:[aclnnEmbedding_1] check embeddingResult != nullptr failed.
[ERROR] OP(1004400,python3):2025-05-08-17:05:30.469.412 [aclnn_embedding.cpp:176][NNOP][~UniqueExecutor][1005106] errno[561102] OpName:[aclnnEmbedding_1] When aclnnEmbeddingGetWorkspaceSize do success, ReleaseTo(executor) should be called before return.
[ERROR] TUNE(1004400,python3):2025-05-08-17:19:14.606.952 [pywrapper.cpp:94][CANNKB][Tid:1004400]"Traceback (most recent call last):
"
[ERROR] TUNE(1004400,python3):2025-05-08-17:19:14.606.963 [pywrapper.cpp:94][CANNKB][Tid:1004400]"  File "/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/tbe/common/repository_manager/interface.py", line 36, in cann_kb_finalize
    def cann_kb_finalize() -> None:
"
[ERROR] TUNE(1004400,python3):2025-05-08-17:19:14.606.967 [pywrapper.cpp:94][CANNKB][Tid:1004400]"KeyboardInterrupt
"
[ERROR] TUNE(1004400,python3):2025-05-08-17:19:14.608.195 [py_interface.cpp:200][CANNKB][Tid:1004400]"Failed to call cann_kb_finalize."
[ERROR] TUNE(1004400,python3):2025-05-08-17:19:14.608.204 [cann_kb_api.cpp:40][CANNKB][Tid:1004400]"Run CannKbFinalize Error!"
[ERROR] TEFUSION(1004400,python3):2025-05-08-17:19:14.608.213 [python_adapter_manager.cc:235]1004400 Finalize call CannKbFinalize failed. res = [6].
```

Ascend/pytorch
暂停

内容风险标识

评论 (3)

Ascend/pytorch暂停 .gitee-modal { width: 500px !important; }

内容风险标识