mindieservice_daemon 启动报错， NPU out of memory （PyTorch）_MindIE_昇腾论坛

# mindieservice_daemon 启动报错， NPU out of memory （PyTorch）_MindIE_昇腾论坛

## 概述

本文档基于昇腾社区论坛帖子生成的技术教程。

**原始链接**: https://www.hiascend.com/forum/thread-0250191041317878336-1-1.html

**生成时间**: 2025-08-27 10:33:56

---

## 问题描述

启动脚本 docker run -it -d --net=host --shm-size=8g \ --privileged \ --name deepseek32b \ --device=/dev/davinci_manager \ --device=/dev/hisi_hdc \ --device=/dev/devmm_svm \ -v /usr/local/Ascend/driver:/usr/local/Ascend/driver:ro \ -v /usr/local/sbin:/usr/local/sbin:ro \ -v /data/datasets:/data/datasets:ro \ mindie:2.0.RC2-300I-Duo-py311-openeuler24.03-lts bash 服务器信息 CPU: 2*Kunpeng 920 32C@2.6G;2.内存:128G; 3.GPU算力:Atlas 300l Pro 24G 推理卡*2; 4. 硬盘:1块内置240G SSD+4块4T硬盘; 5.网卡四口千兆5.电源900W*2 报错信息如下 [2025-08-20 17:33:28.020+0800] [1468] [281472887877984] [batchscheduler] [ERROR] [model.py:62] : [Model] >>> Exception:NPU out of memory. Tried to allocate 272.00 MiB (NPU 1; 21.02 GiB total capacity; 18.88 GiB already allocated; 18.88 GiB current active; 678.39 MiB free; 18.96 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. Traceback (most recent call last): File "/usr/local/lib/python3.11/site-packages/model_wrapper/model...

## 完整内容

启动脚本 docker run -it -d --net=host --shm-size=8g \ --privileged \ --name deepseek32b \ --device=/dev/davinci_manager \ --device=/dev/hisi_hdc \ --device=/dev/devmm_svm \ -v /usr/local/Ascend/driver:/usr/local/Ascend/driver:ro \ -v /usr/local/sbin:/usr/local/sbin:ro \ -v /data/datasets:/data/datasets:ro \ mindie:2.0.RC2-300I-Duo-py311-openeuler24.03-lts bash 服务器信息 CPU: 2*Kunpeng 920 32C@2.6G;2.内存:128G; 3.GPU算力:Atlas 300l Pro 24G 推理卡*2; 4. 硬盘:1块内置240G SSD+4块4T硬盘; 5.网卡四口千兆5.电源900W*2 报错信息如下 [2025-08-20 17:33:28.020+0800] [1468] [281472887877984] [batchscheduler] [ERROR] [model.py:62] : [Model] >>> Exception:NPU out of memory. Tried to allocate 272.00 MiB (NPU 1; 21.02 GiB total capacity; 18.88 GiB already allocated; 18.88 GiB current active; 678.39 MiB free; 18.96 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. Traceback (most recent call last): File "/usr/local/lib/python3.11/site-packages/model_wrapper/model.py", line 60, in initialize return self.python_model.initialize(config)  File "/usr/local/lib/python3.11/site-packages/model_wrapper/standard_model.py", line 133, in initialize self.generator = Generator(  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/generator.py", line 189, in __init__ self.generator_backend = get_generator_backend(model_config)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/__init__.py", line 31, in get_generator_backend return generator_cls(model_config)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_torch.py", line 133, in __init__ super().__init__(model_config) File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_backend.py", line 135, in __init__ self.model_wrapper = get_model_wrapper(model_config, backend_type)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/__init__.py", line 15, in get_model_wrapper return wrapper_cls(**model_config)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/atb/atb_model_wrapper.py", line 79, in __init__ self.model_runner.load_weights() File "/usr/local/Ascend/atb-models/atb_llm/runner/model_runner.py", line 232, in load_weights self.model.to(weights.device) File "/usr/local/lib64/python3.11/site-packages/torch_npu/utils/_module.py", line 78, in to return self._apply(convert)  File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 810, in _apply module._apply(fn) File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 810, in _apply module._apply(fn) File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 810, in _apply module._apply(fn) [Previous line repeated 3 more times] File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 833, in _apply param_applied = fn(param)  File "/usr/local/lib64/python3.11/site-packages/torch_npu/utils/_module.py", line 76, in convert return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)  RuntimeError: NPU out of memory. Tried to allocate 272.00 MiB (NPU 1; 21.02 GiB total capacity; 18.88 GiB already allocated; 18.88 GiB current active; 678.39 MiB free; 18.96 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. [2025-08-20 17:33:28.020+0800] [1467] [281473449849184] [batchscheduler] [ERROR] [model.py:62] : [Model] >>> Exception:NPU out of memory. Tried to allocate 272.00 MiB (NPU 0; 21.02 GiB total capacity; 18.88 GiB already allocated; 18.88 GiB current active; 678.70 MiB free; 18.96 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. Traceback (most recent call last): File "/usr/local/lib/python3.11/site-packages/model_wrapper/model.py", line 60, in initialize return self.python_model.initialize(config)  File "/usr/local/lib/python3.11/site-packages/model_wrapper/standard_model.py", line 133, in initialize self.generator = Generator(  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/generator.py", line 189, in __init__ self.generator_backend = get_generator_backend(model_config)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/__init__.py", line 31, in get_generator_backend return generator_cls(model_config)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_torch.py", line 133, in __init__ super().__init__(model_config) File "/usr/local/lib/python3.11/site-packages/mindie_llm/text_generator/adapter/generator_backend.py", line 135, in __init__ self.model_wrapper = get_model_wrapper(model_config, backend_type)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/__init__.py", line 15, in get_model_wrapper return wrapper_cls(**model_config)  File "/usr/local/lib/python3.11/site-packages/mindie_llm/modeling/model_wrapper/atb/atb_model_wrapper.py", line 79, in __init__ self.model_runner.load_weights() File "/usr/local/Ascend/atb-models/atb_llm/runner/model_runner.py", line 232, in load_weights self.model.to(weights.device) File "/usr/local/lib64/python3.11/site-packages/torch_npu/utils/_module.py", line 78, in to return self._apply(convert)  File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 810, in _apply module._apply(fn) File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 810, in _apply module._apply(fn) File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 810, in _apply module._apply(fn) [Previous line repeated 3 more times] File "/usr/local/lib64/python3.11/site-packages/torch/nn/modules/module.py", line 833, in _apply param_applied = fn(param)  File "/usr/local/lib64/python3.11/site-packages/torch_npu/utils/_module.py", line 76, in convert return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking) 尝试过将npuMemSize为10,也报同样的错请问如何解决 参考的链接 https://www.hiascend.com/forum/thread-02114177941447949034-1-1.html

---

## 技术要点总结

基于以上内容，主要技术要点包括：

1. **问题类型**: 错误处理
2. **涉及技术**: MindIE, HTTPS, PyTorch, GPU, NPU, Docker, Atlas, 昇腾
3. **解决方案**: 请参考完整内容中的解决方案

## 相关资源

- 昇腾社区: https://www.hiascend.com/
- 昇腾论坛: https://www.hiascend.com/forum/

---

*本文档由AI自动生成，仅供参考。如有疑问，请参考原始帖子。*