rwkv_cuda

simple minimal dependency test

layernorm / softmax / wkv_forward using oneflow style custom cuda kernel

gemm / gemv using slightly modified cutlass 3.1

argsort using thrust

minimal test: under bin folder invoked with nodejs and koffi

Name		Name	Last commit message	Last commit date
Latest commit History 11 Commits
bin		bin
cutlass		cutlass
kernel		kernel
lib_rwkv		lib_rwkv
LICENSE		LICENSE
README.md		README.md
convert_safetensors.py		convert_safetensors.py
json.hpp		json.hpp
main.cpp		main.cpp
mmap.hpp		mmap.hpp
rwkv.cuh		rwkv.cuh
rwkv_cuda.sln		rwkv_cuda.sln
rwkv_cuda.vcxproj		rwkv_cuda.vcxproj
rwkv_cuda.vcxproj.filters		rwkv_cuda.vcxproj.filters
rwkv_model.hpp		rwkv_model.hpp
safe_tensors.hpp		safe_tensors.hpp
sampler.hpp		sampler.hpp
simd.cpp		simd.cpp
simd.hpp		simd.hpp
thrust.cpp		thrust.cpp
thrust.hpp		thrust.hpp
tokenizer.hpp		tokenizer.hpp
util.hpp		util.hpp

Provide feedback