search

k2-fsa / k2

FSA/FST algorithms, differentiable, with PyTorch compatibility.
https://k2-fsa.github.io/k2
Apache License 2.0
1.11k stars 213 forks source link

Failed to build k2 from source due to an error in deserialization.cu #1176

Open josh6688 opened 1 year ago

josh6688 commented 1 year ago

I tried to build k2 inside a docker image provided by nvidia. (nvcr.io/nvidia/pytorch:21.06-py3)

PyTorch version: 1.9.0a0+c3d40fd
PyTorch cuda version: 11.3

Build command was python setup.py install The error below occurred when compiling deserialization.cu

/workspace/k2/k2/torch/csrc/deserialization.cu(404): error: no suitable constructor exists to convert from "const char [1]" to "c10::optional<torch::jit::TypeResolver>"

/workspace/k2/k2/torch/csrc/deserialization.cu(404): error: no suitable constructor exists to convert from "const char [1]" to "c10::optional<torch::jit::ObjLoader>"

/workspace/k2/k2/torch/csrc/deserialization.cu(404): error: no suitable user-defined conversion from "lambda [](const c10::QualifiedName &)->c10::StrongTypePtr" to "c10::optional<c10::Device>" exists

/workspace/k2/k2/torch/csrc/deserialization.cu(404): error: a reference of type "caffe2::serialize::PyTorchStreamReader &" (not const-qualified) cannot be initialized with a value of type "lambda [](c10::StrongTypePtr, c10::IValue)->c10::intrusive_ptr<c10::ivalue::Object, c10::detail::intrusive_target_default_null_type<c10::ivalue::Object>>"

/workspace/k2/k2/torch/csrc/deserialization.cu(405): error: too many arguments in function call