[TTS]Out of memory with fastspeech2_csmsc

[DBDXSS]

参考https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts3
运行代码，环境配置如下：
python3 -m pip install paddlepaddle-gpu==2.5.2.post117 paddlespeech -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html

显卡配置：H800 80G cuda 11.8

Building prefix dict from the default dictionary ...
[2025-04-16 14:33:00,898] [ DEBUG] init.py:113 - Building prefix dict from the default dictionary ...
Loading model from cache /tmp/jieba.cache
[2025-04-16 14:33:00,899] [ DEBUG] init.py:132 - Loading model from cache /tmp/jieba.cache
Loading model cost 0.505 seconds.
[2025-04-16 14:33:01,403] [ DEBUG] init.py:164 - Loading model cost 0.505 seconds.
Prefix dict has been built successfully.
[2025-04-16 14:33:01,403] [ DEBUG] init.py:166 - Prefix dict has been built successfully.
W0416 14:33:12.098174 5426 interpreter_util.cc:838] range raises an exception paddle::memory::allocation::BadAlloc,

---

C++ Traceback (most recent call last):

0 run_program_ad_func(std::vector<paddle::Tensor, std::allocatorpaddle::Tensor > const&, std::vector<paddle::Tensor, std::allocatorpaddle::Tensor > const&, std::vector<paddle::Tensor*, std::allocatorpaddle::Tensor* >&, std::vector<paddle::framework::Scope*, std::allocatorpaddle::framework::Scope* >&, std::vector<paddle::Tensor*, std::allocatorpaddle::Tensor* >&, paddle::framework::AttributeMap const&)
1 RunProgramAPI(std::vector<paddle::Tensor, std::allocatorpaddle::Tensor > const&, std::vector<paddle::Tensor, std::allocatorpaddle::Tensor > const&, std::vector<paddle::Tensor*, std::allocatorpaddle::Tensor* >&, std::vector<paddle::framework::Scope*, std::allocatorpaddle::framework::Scope* >&, std::vector<paddle::Tensor*, std::allocatorpaddle::Tensor* >&, bool, paddle::framework::AttributeMap const&)
2 paddle::framework::InterpreterCore::Run(std::vector<std::string, std::allocator<std::string > > const&, bool)
3 paddle::framework::interpreter::BuildOpFuncList(phi::Place const&, paddle::framework::BlockDesc const&, std::set<std::string, std::less<std::string >, std::allocator<std::string > > const&, std::vector<paddle::framework::OpFuncNode, std::allocatorpaddle::framework::OpFuncNode >, paddle::framework::VariableScope, paddle::framework::interpreter::ExecutionConfig const&, bool, bool)
4 void phi::ArangeKernel<long, phi::GPUContext>(phi::GPUContext const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor const&, phi::DenseTensor*)
5 long* phi::DeviceContext::Alloc(phi::TensorBase*, unsigned long, bool) const
6 phi::DeviceContext::Impl::Alloc(phi::TensorBase*, phi::Place const&, phi::DataType, unsigned long, bool, bool) const
7 phi::DenseTensor::AllocateFrom(phi::Allocator*, phi::DataType, unsigned long, bool)
8 paddle::memory::allocation::Allocator::Allocate(unsigned long)
9 paddle::memory::allocation::StatAllocator::AllocateImpl(unsigned long)
10 paddle::memory::allocation::Allocator::Allocate(unsigned long)
11 paddle::memory::allocation::Allocator::Allocate(unsigned long)
12 paddle::memory::allocation::Allocator::Allocate(unsigned long)
13 paddle::memory::allocation::CUDAAllocator::AllocateImpl(unsigned long)
14 std::string phi::enforce::GetCompleteTraceBackString<std::string >(std::string&&, char const*, int)
15 phi::enforce::GetCurrentTraceBackStringabi:cxx11

---

Error Message Summary:

ResourceExhaustedError:

Out of memory error on GPU 0. Cannot allocate 7.019809EB memory on GPU 0, 915.187500MB memory has been allocated and available memory is only 78.426208GB.

显示需要7.019809EB，修改batchsize无效

[DBDXSS]

使用cpu可以运行，设置gpu如上

python3 ${BIN_DIR}/../synthesize_e2e.py
--am=fastspeech2_csmsc
--am_config=fastspeech2_nosil_baker_ckpt_0.4/default.yaml
--am_ckpt=fastspeech2_nosil_baker_ckpt_0.4/snapshot_iter_76000.pdz
--am_stat=fastspeech2_nosil_baker_ckpt_0.4/speech_stats.npy
--voc=pwgan_csmsc
--voc_config=pwg_baker_ckpt_0.4/pwg_default.yaml
--voc_ckpt=pwg_baker_ckpt_0.4/pwg_snapshot_iter_400000.pdz
--voc_stat=pwg_baker_ckpt_0.4/pwg_stats.npy
--lang=zh
--text=${BIN_DIR}/../../assets/sentences.txt
--output_dir=exp/default/test_e2e
--inference_dir=exp/default/inference
--phones_dict=fastspeech2_nosil_baker_ckpt_0.4/phone_id_map.txt
--ngpu=0

[zxcd]

可以在保证gpu显存的情况下尝试使用paddlespeech == develop + paddlepaddle-gpu>=3.0beta2 看看是否能跑

[DBDXSS]

可以在保证gpu显存的情况下尝试使用paddlespeech == develop + paddlepaddle-gpu>=3.0beta2 看看是否能跑

请问有可参考的下载文档吗

[DBDXSS]

可以在保证gpu显存的情况下尝试使用paddlespeech == develop + paddlepaddle-gpu>=3.0beta2 看看是否能跑

我测试了下都安装了，输出下面错误，InvalidArgumentError: The 3-th dimension of input tensor (17) must match or be broadcastable to the corresponding dimension (4352512999702696242) in shape. 似乎这也是为什么会爆显存的原因。

[zxcd]

${BIN_DIR}/../../assets/sentences.txt 中的文本是默认的文本吗？