add pre-compiled beta cuda kernel (rwkv-beta==0.8.5, 40%+ faster for fp16) (thanks to #180, pre-compiled kernel of RTX 40 Series will be included later)

.github/workflows/release.yml

@@ -108,6 +108,7 @@ jobs:
           cd ..
           go install github.com/wailsapp/wails/v2/cmd/wails@latest
           rm -rf ./backend-python/wkv_cuda_utils
+          rm ./backend-python/rwkv_pip/beta/wkv_cuda.pyd
           rm ./backend-python/get-pip.py
           sed -i '1,2d' ./backend-golang/wsl_not_windows.go
           rm ./backend-golang/wsl.go
@@ -140,6 +141,7 @@ jobs:
           cd ..
           go install github.com/wailsapp/wails/v2/cmd/wails@latest
           rm -rf ./backend-python/wkv_cuda_utils
+          rm ./backend-python/rwkv_pip/beta/wkv_cuda.pyd
           rm ./backend-python/get-pip.py
           sed -i '' '1,2d' ./backend-golang/wsl_not_windows.go
           rm ./backend-golang/wsl.go

backend-python/rwkv_pip/beta/model.py

@@ -94,6 +94,7 @@ if os.environ.get("RWKV_CUDA_ON") == "1":
                 f"{current_path}/cuda/att_one_v5.cu",
             ],
             verbose=True,
+            extra_ldflags=["cublas.lib"],
             extra_cuda_cflags=[
                 "-t 4",
                 "-std=c++17",