Killian Q. Zhuo

可以支持，不过我目前时间不多，我会慢慢增加。

Yes, this is a known issue, please send a pull request if you fixed it, many thanks!

Sorry that you'd better to do some workaround at this time, I'm too busy on other projects to fix this now.

Some PDF files need external fonts to render, and WASM in browsers doesn't have the ability to access native files. I have managed to produce a version which will fetch...

@yebai define ```diff +tensor_from_ptr(p::Ptr{Tensor{T, N}}) where {T, N} = Tensor{T, N}(p, nothing) ``` and ```diff -function grad(self::Tensor) +function grad(self::Tensor{T, N}) where {T, N} outputs__ = Int[0] __cret = ccall((:atg_grad, :libtorch_capi),...

``` c++ #include #include #include #include #include int main() { torch::Tensor t = torch::rand({10000}, torch::requires_grad(1)); torch::Tensor r(t[0].clone()); r.set_requires_grad(1); for(int i=0; i< 10000; i++ ) { r += t[i]; } r.backward();...

No, on linux, it's a typo, it's an ELF, sorry for the confusion. ``` $ file example-app example-app: ELF 64-bit LSB shared object, x86-64, version 1 (GNU/Linux), dynamically linked, interpreter...

``` julia> @btime ThArrays.gradient(mysum, x); 1.257 s (188995 allocations: 7.92 MiB) Julia> $ time ./example-app >/dev/null real 0m0.519s user 0m0.491s sys 0m0.013s ```

In real-world practice, we may not do indexing in such a frequency, e.g. we should use `sum` instead of `mysum`. About torchscript, I didn't mean a source to source translation,...

I did some optimizations on ThArrays, https://github.com/TuringLang/ThArrays.jl/commit/c0217179fcc99e237e04ccb9161c2d0c3fc96003, indexing is a little faster(10%?) now.