Blealtan

FYI C++ lambda expressions are just closure objects and require no low-level support at no cost; in contrast, coroutines in C++20 require only several intrinsic functions that allow control flow...

AFAIK Python has GIL such that no Python code can be interpreted simultaneously in different threads. Are you sure the code is actually running in multi-thread? Is it possible to...

Same issue here. Also, even in native `cmd.exe` or `PowerShell.exe` mouse events are working correctly now.

Another problem is about the shadows; when the Quake window is opening, the shadow below the bottom border is shown up before the bottom actually extends there. The below screenshot...

In my case, `LD_LIBRARY_PATH=gcc-c-api ./gcc-with-python ...` would load the plugin correctly.

TUNA-SIGCHEN？

@YovaKem Would you mind try change this ```python # The CUDA kernel will fill those tensors. g_time_decay = torch.empty_like( time_decay, memory_format=torch.contiguous_format, dtype=torch.bfloat16 if input_dtype == torch.bfloat16 else torch.float32, ) g_time_first...

@YovaKem AFAIK, `with_state` is used only in inference now (in existing non-`transformers` implementations throughout the RWKV community). However, with proper implementation, this will allow more efficient training on long sequences,...

Actually DllImport SHOULD work fine in any implementation that supports .Net Standard, always. It's a required interface, and should work with anything like a shared object or a DLL; in...

Sorry but I've forget the global variables.. It seems that being totally platform independent is impossible due to this design of original HDF5 library. However, if possible, distribute runtime packages...