torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 14.00 MiB (GPU 0; 4.00 GiB total capacity; 3.21 G

这个错误表示在分配CUDA内存时出现了问题，因为GPU的内存不足。你可以尝试以下几种方法来解决这个问题：

1. 减少模型的大小：如果你使用的模型比较大，可以尝试使用更小的模型或减少模型的参数数量。

2. 减少批次大小：减少每个批次中样本的数量，以减少内存的使用量。

3. 释放不必要的内存：在每个批次处理完后，通过调用`torch.cuda.empty_cache()`释放不必要的内存。

4. 使用更高容量的GPU：如果你有更高容量的GPU可用，可以尝试切换到使用该GPU。

5. 使用分布式训练：如果你有多个GPU可用，可以尝试使用分布式训练将模型参数分布到多个GPU上，以减少单个GPU上的内存使用。

请注意，如果你的代码在CPU上运行正常，但在GPU上出现内存问题，那么很可能是由于GPU的内存限制导致的。

相关问题

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 148.00 MiB (GPU 0; 4.00 GiB total capacity; 5.23 GiB already allocated;

torch.cuda.OutOfMemoryError是指在使用PyTorch时，尝试在CUDA显存中分配内存时出现错误，因为显存已经被其他操作占用完毕。其中引用和引用提到了相同的错误信息和可能的解决方法。根据这些引用内容，可以推测解决此错误的方法是通过设置max_split_size_mb参数来避免内存碎片化。你可以参考PyTorch的Memory Management和PYTORCH_CUDA_ALLOC_CONF文档了解更多的信息。引用也提到了类似的错误信息，但给出了不同的显存容量和已分配内存的数值。这说明出现该错误的具体原因可能因系统配置不同而有所不同。<span class="em">1</span><span class="em">2</span><span class="em">3</span>
#### 引用[.reference_title]
- *1* *2* [成功解决torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 86.00 MiB (GPU 0； 2.00 ...](https://blog.csdn.net/qq_41185868/article/details/131015210)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v93^chatsearchT3_1"}}] [.reference_item style="max-width: 50%"]
- *3* [报错记录torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 90.00 MiB (GPU 0； 7.93 ...](https://blog.csdn.net/gongkeguo/article/details/127674524)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v93^chatsearchT3_1"}}] [.reference_item style="max-width: 50%"]
[ .reference_list ]

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 2.00 GiB total capacity; 1.67 GiB a

vailable; 40.00 KiB already allocated; 19.94 MiB free; 4.00 MiB cached)

This error message indicates that the program tried to allocate 20.00 MiB of memory on the GPU, but there was not enough memory available. The GPU has a total capacity of 2.00 GiB, but only 1.67 GiB was available at the time. 40.00 KiB had already been allocated, leaving only 19.94 MiB free. Additionally, 4.00 MiB had been cached, which means that it was reserved for future use but not currently in use.

To resolve this issue, you can try one or more of the following:

1. Reduce the size of the input data or the size of the model being used. This will reduce the amount of memory required.

2. Increase the GPU memory capacity. If possible, add more memory to the GPU or switch to a GPU with a larger memory capacity.

3. Use a smaller batch size. This will reduce the amount of memory required for each iteration.

4. Use gradient checkpointing. This technique allows the model to compute gradients for small subsets of parameters at a time, reducing the amount of memory required.

5. Use mixed precision training. This technique allows the model to use 16-bit floating point numbers instead of 32-bit, reducing the amount of memory required.