Merge pull request #1128 from akx/type-fixes

Minor type/doc fixes

Author: Titus-von-Koeller

bitsandbytes/functional.py

@@ -1618,18 +1618,18 @@ def optimizer_update_8bit(
     g: Tensor,
     p: Tensor,
     state1: Tensor,
-    state2: Tensor,
+    state2: Optional[torch.Tensor],
     beta1: float,
     beta2: float,
     eps: float,
     step: int,
     lr: float,
     qmap1: Tensor,
-    qmap2: Tensor,
+    qmap2: Optional[torch.Tensor],
     max1: Tensor,
-    max2: Tensor,
+    max2: Optional[torch.Tensor],
     new_max1: Tensor,
-    new_max2: Tensor,
+    new_max2: Optional[torch.Tensor],
     weight_decay: float = 0.0,
     gnorm_scale: float = 1.0,
     unorm_vec: Optional[torch.Tensor] = None,
@@ -1751,16 +1751,16 @@ def optimizer_update_8bit_blockwise(
     g: Tensor,
     p: Tensor,
     state1: Tensor,
-    state2: Tensor,
+    state2: Optional[torch.Tensor],
     beta1: float,
     beta2: float,
     eps: float,
     step: int,
     lr: float,
     qmap1: Tensor,
-    qmap2: Tensor,
+    qmap2: Optional[torch.Tensor],
     absmax1: Tensor,
-    absmax2: Tensor,
+    absmax2: Optional[torch.Tensor],
     weight_decay: float = 0.0,
     gnorm_scale: float = 1.0,
     skip_zeros=False,

bitsandbytes/nn/modules.py

@@ -658,8 +658,8 @@ class Linear8bitLt(nn.Linear):
 
     def __init__(
         self,
-        input_features,
-        output_features,
+        input_features: int,
+        output_features: int,
         bias=True,
         has_fp16_weights=True,
         memory_efficient_backward=False,
@@ -671,9 +671,9 @@ def __init__(
         Initialize Linear8bitLt class.
 
         Args:
-            input_features (`str`):
+            input_features (`int`):
                 Number of input features of the linear layer.
-            output_features (`str`):
+            output_features (`int`):
                 Number of output features of the linear layer.
             bias (`bool`, defaults to `True`):
                 Whether the linear class uses the bias term as well.

bitsandbytes/optim/adagrad.py

@@ -38,8 +38,8 @@ def __init__(
                 The epsilon value prevents division by zero in the optimizer.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -105,8 +105,8 @@ def __init__(
                 The epsilon value prevents division by zero in the optimizer.
             optim_bits (`int`, defaults to 8):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -173,8 +173,8 @@ def __init__(
                 The epsilon value prevents division by zero in the optimizer.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):

bitsandbytes/optim/adam.py

@@ -47,8 +47,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -108,8 +108,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -169,8 +169,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -230,8 +230,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -291,8 +291,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -352,8 +352,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):

bitsandbytes/optim/adamw.py

@@ -39,8 +39,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -100,8 +100,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -161,8 +161,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -221,8 +221,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -281,8 +281,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -341,8 +341,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):

bitsandbytes/optim/lamb.py

@@ -45,8 +45,8 @@ def __init__(
                 Whether to use the AdamW variant.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -109,8 +109,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             adam_w_mode (`bool`, defaults to `True`):
                 Whether to use the AdamW variant.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -173,8 +173,8 @@ def __init__(
                 Whether to use the [AMSGrad](https://hf.co/papers/1904.09237) variant of Adam that uses the maximum of past squared gradients instead.
             adam_w_mode (`bool`, defaults to `True`):
                 Whether to use the AdamW variant.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):

bitsandbytes/optim/lars.py

@@ -41,8 +41,8 @@ def __init__(
                 Whether to use Nesterov momentum.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -98,8 +98,8 @@ def __init__(
                 The weight decay value for the optimizer.
             nesterov (`bool`, defaults to `False`):
                 Whether to use Nesterov momentum.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -155,8 +155,8 @@ def __init__(
                 The weight decay value for the optimizer.
             nesterov (`bool`, defaults to `False`):
                 Whether to use Nesterov momentum.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):

bitsandbytes/optim/lion.py

@@ -33,8 +33,8 @@ def __init__(
                 The weight decay value for the optimizer.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -85,8 +85,8 @@ def __init__(
                 The beta values are the decay rates of the first and second-order moment of the optimizer.
             weight_decay (`float`, defaults to 0):
                 The weight decay value for the optimizer.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -137,8 +137,8 @@ def __init__(
                 The beta values are the decay rates of the first and second-order moment of the optimizer.
             weight_decay (`float`, defaults to 0):
                 The weight decay value for the optimizer.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -191,8 +191,8 @@ def __init__(
                 The weight decay value for the optimizer.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -242,8 +242,8 @@ def __init__(
                 The weight decay value for the optimizer.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):
@@ -293,8 +293,8 @@ def __init__(
                 The weight decay value for the optimizer.
             optim_bits (`int`, defaults to 32):
                 The number of bits of the optimizer state.
-            args (`dict`, defaults to `None`):
-                A dictionary with additional arguments.
+            args (`object`, defaults to `None`):
+                An object with additional arguments.
             min_8bit_size (`int`, defaults to 4096):
                 The minimum number of elements of the parameter tensors for 8-bit optimization.
             percentile_clipping (`int`, defaults to 100):