Simplify statements into equivalent, modern variants

via pyupgrade --py37-plus. The changes e.g. are subclassing from object, calling super() with super(ThisClass, self), or old-style syntax formatting.

Author: tomaarsen

bitsandbytes/autograd/_functions.py

@@ -18,7 +18,7 @@ def prod(iterable):
     This is particularly important for small models where outlier features
     are less systematic and occur with low frequency.
 """
-class GlobalOutlierPooler(object):
+class GlobalOutlierPooler:
     _instance = None
 
     def __init__(self):

bitsandbytes/cextension.py

@@ -5,7 +5,7 @@
 from warnings import warn
 
 
-class CUDASetup(object):
+class CUDASetup:
     _instance = None
 
     def __init__(self):

bitsandbytes/cuda_setup/main.py

@@ -127,7 +127,7 @@ def evaluate_cuda_setup():
         cuda_setup.add_log_entry("WARNING: No libcudart.so found! Install CUDA or the cudatoolkit package (anaconda)!", is_warning=True)
         return binary_name
 
-    cuda_setup.add_log_entry((f"CUDA SETUP: CUDA runtime path found: {cudart_path}"))
+    cuda_setup.add_log_entry(f"CUDA SETUP: CUDA runtime path found: {cudart_path}")
     cuda = get_cuda_lib_handle()
     cc = get_compute_capability(cuda)
     cuda_setup.add_log_entry(f"CUDA SETUP: Highest compute capability among GPUs detected: {cc}")

bitsandbytes/functional.py

@@ -82,7 +82,7 @@ def prod(iterable):
     )
 
 
-class CUBLAS_Context(object):
+class CUBLAS_Context:
     _instance = None
 
     def __init__(self):
@@ -112,7 +112,7 @@ def get_context(self, device):
         return self.context[device.index]
 
 
-class Cusparse_Context(object):
+class Cusparse_Context:
     _instance = None
 
     def __init__(self):
@@ -1417,7 +1417,7 @@ def get_colrow_absmax(
     return row_stats, col_stats, nnz_block_ptr
 
 
-class COOSparseTensor(object):
+class COOSparseTensor:
     def __init__(self, rows, cols, nnz, rowidx, colidx, values):
         assert rowidx.dtype == torch.int32
         assert colidx.dtype == torch.int32
@@ -1434,7 +1434,7 @@ def __init__(self, rows, cols, nnz, rowidx, colidx, values):
         self.values = values
 
 
-class CSRSparseTensor(object):
+class CSRSparseTensor:
     def __init__(self, rows, cols, nnz, rowptr, colidx, values):
         assert rowptr.dtype == torch.int32
         assert colidx.dtype == torch.int32
@@ -1451,7 +1451,7 @@ def __init__(self, rows, cols, nnz, rowptr, colidx, values):
         self.values = values
 
 
-class CSCSparseTensor(object):
+class CSCSparseTensor:
     def __init__(self, rows, cols, nnz, colptr, rowidx, values):
         assert colptr.dtype == torch.int32
         assert rowidx.dtype == torch.int32

bitsandbytes/nn/modules.py

@@ -39,7 +39,7 @@ def __init__(
         sparse: bool = False,
         _weight: Optional[Tensor] = None,
     ) -> None:
-        super(StableEmbedding, self).__init__(
+        super().__init__(
             num_embeddings,
             embedding_dim,
             padding_idx,
@@ -96,7 +96,7 @@ def __init__(
         sparse: bool = False,
         _weight: Optional[Tensor] = None,
     ) -> None:
-        super(Embedding, self).__init__(
+        super().__init__(
             num_embeddings,
             embedding_dim,
             padding_idx,
@@ -225,7 +225,7 @@ def __init__(
         threshold=0.0,
         index=None,
     ):
-        super(Linear8bitLt, self).__init__(
+        super().__init__(
             input_features, output_features, bias
         )
         self.state = bnb.MatmulLtState()

bitsandbytes/optim/adagrad.py

@@ -21,18 +21,18 @@ def __init__(
         block_wise=True,
     ):
         if not 0.0 <= lr:
-            raise ValueError("Invalid learning rate: {}".format(lr))
+            raise ValueError(f"Invalid learning rate: {lr}")
         if not 0.0 <= weight_decay:
             raise ValueError(
-                "Invalid weight_decay value: {}".format(weight_decay)
+                f"Invalid weight_decay value: {weight_decay}"
             )
         if not 0.0 <= eps:
-            raise ValueError("Invalid epsilon value: {}".format(eps))
+            raise ValueError(f"Invalid epsilon value: {eps}")
         if initial_accumulator_value != 0.0:
             raise ValueError("Initial accumulator value != 0.0 not supported!")
         if lr_decay != 0.0:
             raise ValueError("Lr Decay != 0.0 not supported!")
-        super(Adagrad, self).__init__(
+        super().__init__(
             "adagrad",
             params,
             lr,
@@ -63,19 +63,19 @@ def __init__(
         block_wise=True,
     ):
         if not 0.0 <= lr:
-            raise ValueError("Invalid learning rate: {}".format(lr))
+            raise ValueError(f"Invalid learning rate: {lr}")
         if not 0.0 <= weight_decay:
             raise ValueError(
-                "Invalid weight_decay value: {}".format(weight_decay)
+                f"Invalid weight_decay value: {weight_decay}"
             )
         if not 0.0 <= eps:
-            raise ValueError("Invalid epsilon value: {}".format(eps))
+            raise ValueError(f"Invalid epsilon value: {eps}")
         if initial_accumulator_value != 0.0:
             raise ValueError("Initial accumulator value != 0.0 not supported!")
         if lr_decay != 0.0:
             raise ValueError("Lr Decay != 0.0 not supported!")
         assert block_wise
-        super(Adagrad8bit, self).__init__(
+        super().__init__(
             "adagrad",
             params,
             lr,
@@ -106,18 +106,18 @@ def __init__(
         block_wise=True,
     ):
         if not 0.0 <= lr:
-            raise ValueError("Invalid learning rate: {}".format(lr))
+            raise ValueError(f"Invalid learning rate: {lr}")
         if not 0.0 <= weight_decay:
             raise ValueError(
-                "Invalid weight_decay value: {}".format(weight_decay)
+                f"Invalid weight_decay value: {weight_decay}"
             )
         if not 0.0 <= eps:
-            raise ValueError("Invalid epsilon value: {}".format(eps))
+            raise ValueError(f"Invalid epsilon value: {eps}")
         if initial_accumulator_value != 0.0:
             raise ValueError("Initial accumulator value != 0.0 not supported!")
         if lr_decay != 0.0:
             raise ValueError("Lr Decay != 0.0 not supported!")
-        super(Adagrad32bit, self).__init__(
+        super().__init__(
             "adagrad",
             params,
             lr,

bitsandbytes/optim/adam.py

@@ -28,7 +28,7 @@ def __init__(
         percentile_clipping=100,
         block_wise=True,
     ):
-        super(Adam, self).__init__(
+        super().__init__(
             "adam",
             params,
             lr,
@@ -57,7 +57,7 @@ def __init__(
         percentile_clipping=100,
         block_wise=True,
     ):
-        super(Adam8bit, self).__init__(
+        super().__init__(
             "adam",
             params,
             lr,
@@ -86,7 +86,7 @@ def __init__(
         percentile_clipping=100,
         block_wise=True,
     ):
-        super(Adam32bit, self).__init__(
+        super().__init__(
             "adam",
             params,
             lr,
@@ -146,7 +146,7 @@ def __init__(
             weight_decay=weight_decay,
             amsgrad=amsgrad,
         )
-        super(AnalysisAdam, self).__init__(params, defaults)
+        super().__init__(params, defaults)
         self.analysis = bnb_analysis
         self.savedir = savedir
 

bitsandbytes/optim/adamw.py

@@ -20,7 +20,7 @@ def __init__(
         percentile_clipping=100,
         block_wise=True,
     ):
-        super(AdamW, self).__init__(
+        super().__init__(
             "adam",
             params,
             lr,
@@ -49,7 +49,7 @@ def __init__(
         percentile_clipping=100,
         block_wise=True,
     ):
-        super(AdamW8bit, self).__init__(
+        super().__init__(
             "adam",
             params,
             lr,
@@ -78,7 +78,7 @@ def __init__(
         percentile_clipping=100,
         block_wise=True,
     ):
-        super(AdamW32bit, self).__init__(
+        super().__init__(
             "adam",
             params,
             lr,

bitsandbytes/optim/lamb.py

@@ -23,7 +23,7 @@ def __init__(
         block_wise=False,
         max_unorm=1.0,
     ):
-        super(LAMB, self).__init__(
+        super().__init__(
             "lamb",
             params,
             lr,
@@ -56,7 +56,7 @@ def __init__(
         block_wise=False,
         max_unorm=1.0,
     ):
-        super(LAMB8bit, self).__init__(
+        super().__init__(
             "lamb",
             params,
             lr,
@@ -89,7 +89,7 @@ def __init__(
         block_wise=False,
         max_unorm=1.0,
     ):
-        super(LAMB32bit, self).__init__(
+        super().__init__(
             "lamb",
             params,
             lr,

bitsandbytes/optim/lars.py

@@ -27,7 +27,7 @@ def __init__(
             raise NotImplementedError(
                 f"LARS without momentum is not supported!"
             )
-        super(LARS, self).__init__(
+        super().__init__(
             "lars",
             params,
             lr,
@@ -61,7 +61,7 @@ def __init__(
             raise NotImplementedError(
                 f"LARS without momentum is not supported!"
             )
-        super(LARS8bit, self).__init__(
+        super().__init__(
             "lars",
             params,
             lr,
@@ -95,7 +95,7 @@ def __init__(
             raise NotImplementedError(
                 f"LARS without momentum is not supported!"
             )
-        super(LARS32bit, self).__init__(
+        super().__init__(
             "lars",
             params,
             lr,
@@ -123,12 +123,12 @@ def __init__(
         max_unorm=0.02,
     ):
         if lr < 0.0:
-            raise ValueError("Invalid learning rate: {}".format(lr))
+            raise ValueError(f"Invalid learning rate: {lr}")
         if momentum < 0.0:
-            raise ValueError("Invalid momentum value: {}".format(momentum))
+            raise ValueError(f"Invalid momentum value: {momentum}")
         if weight_decay < 0.0:
             raise ValueError(
-                "Invalid weight_decay value: {}".format(weight_decay)
+                f"Invalid weight_decay value: {weight_decay}"
             )
 
         defaults = dict(
@@ -143,10 +143,10 @@ def __init__(
             raise ValueError(
                 "Nesterov momentum requires a momentum and zero dampening"
             )
-        super(PytorchLARS, self).__init__(params, defaults)
+        super().__init__(params, defaults)
 
     def __setstate__(self, state):
-        super(PytorchLARS, self).__setstate__(state)
+        super().__setstate__(state)
         for group in self.param_groups:
             group.setdefault("nesterov", False)