Merge branch 'main' into ruff-fixes

.git-blame-ignore-revs

@@ -0,0 +1,8 @@
+# ran black and isort for coherent code formatting
+bfa0e33294f2b1dc25e65a33be2397f989824298
+
+# reran black with linelength 80 for greater readability
+ea7c14f8ef64924f2d0ff80df3cdabf2c7299848
+
+# Remove f-prefix from strings that don't use formatting
+7727fa4c8c6c1ef2b109120aff4196a0a6bf3ed6

bitsandbytes/autograd/_functions.py

@@ -14,9 +14,6 @@
 def prod(iterable):
     return reduce(operator.mul, iterable, 1)
 
-tensor = torch.Tensor
-
-
 # The inverse transformation for the colTuring and colAmpere format were contributed by Alex Borzunov:
 # https://github.com/bigscience-workshop/petals/blob/main/src/petals/utils/linear8bitlt_patch.py
 
@@ -56,7 +53,10 @@ def get_current_outlier_idx(self):
         return torch.Tensor(list(self.outliers)).to(torch.int64)
 
 
-def get_inverse_transform_indices(transform_tile: callable, tile_size: Tuple[int, int]):
+def get_inverse_transform_indices(
+    transform_tile: Callable[[torch.Tensor], torch.Tensor],
+    tile_size: Tuple[int, int],
+):
     """
     Compute a permutation of indices that invert the specified (tiled) matrix transformation
 
@@ -496,7 +496,7 @@ class MatMul4Bit(torch.autograd.Function):
     # backward is mostly the same, but adds one extra clause (see "elif state.CxB is not None")
 
     @staticmethod
-    def forward(ctx, A, B, out=None, bias=None, quant_state: F.QuantState = None):
+    def forward(ctx, A, B, out=None, bias=None, quant_state: Optional[F.QuantState] = None):
         # default of pytorch behavior if inputs are empty
         ctx.is_empty = False
         if prod(A.shape) == 0:
@@ -549,10 +549,10 @@ def backward(ctx, grad_output):
 
 
 def matmul(
-    A: tensor,
-    B: tensor,
-    out: tensor = None,
-    state: MatmulLtState = None,
+    A: torch.Tensor,
+    B: torch.Tensor,
+    out: Optional[torch.Tensor] = None,
+    state: Optional[MatmulLtState] = None,
     threshold=0.0,
     bias=None
 ):
@@ -562,7 +562,7 @@ def matmul(
     return MatMul8bitLt.apply(A, B, out, bias, state)
 
 
-def matmul_4bit(A: tensor, B: tensor, quant_state: F.QuantState, out: tensor = None, bias=None):
+def matmul_4bit(A: torch.Tensor, B: torch.Tensor, quant_state: F.QuantState, out: Optional[torch.Tensor] = None, bias=None):
     assert quant_state is not None
     if A.numel() == A.shape[-1] and A.requires_grad == False:
         if A.shape[-1] % quant_state.blocksize != 0:

bitsandbytes/cuda_setup/main.py

@@ -34,9 +34,9 @@
 # not sure if libcudart.so.12.0 exists in pytorch installs, but it does not hurt
 system = platform.system()
 if system == 'Windows':
-    CUDA_RUNTIME_LIBS: list = ["nvcuda.dll"]
+    CUDA_RUNTIME_LIBS = ["nvcuda.dll"]
 else: # Linux or other
-    CUDA_RUNTIME_LIBS: list = ["libcudart.so", 'libcudart.so.11.0', 'libcudart.so.12.0', 'libcudart.so.12.1', 'libcudart.so.12.2']
+    CUDA_RUNTIME_LIBS = ["libcudart.so", 'libcudart.so.11.0', 'libcudart.so.12.0', 'libcudart.so.12.1', 'libcudart.so.12.2']
 
 # this is a order list of backup paths to search CUDA in, if it cannot be found in the main environmental paths
 backup_paths = []