Fixed default args

bitsandbytes/_ops.py

@@ -352,7 +352,7 @@ if ipex_cpu or ipex_xpu:
 
 torch.library.define(
     "bitsandbytes::optimizer_update_32bit",
-    "(str optimizer_name, Tensor(a0!) g, Tensor(a1!) p, Tensor(a2!) state1, Tensor(a3!)? state2, Tensor(a4!)? unorm_vec, float max_unorm, float param_norm, float beta1, float beta2, float beta3, float alpha, float eps, float weight_decay, int step, float lr, float gnorm_scale, bool skip_zeros) -> ()",
+    "(str optimizer_name, Tensor(a0!) g, Tensor(a1!) p, Tensor(a2!) state1, Tensor(a3!)? state2, Tensor(a4!)? unorm_vec, float max_unorm, float param_norm, float beta1, float beta2, float beta3, float alpha, float eps, float weight_decay, int step, float lr, float gnorm_scale, bool skip_zeros=False) -> ()",
 )
 
 
@@ -395,7 +395,7 @@ def _(
 
 torch.library.define(
     "bitsandbytes::optimizer_update_8bit_blockwise",
-    "(str optimizer_name, Tensor(a0!) g, Tensor(a1!) p, Tensor(a2!) state1, Tensor(a3!)? state2, float beta1, float beta2, float beta3, float alpha, float eps, int step, float lr, Tensor(a4!) qmap1, Tensor(a5!)? qmap2, Tensor(a6!) absmax1, Tensor(a7!)? absmax2, float weight_decay, float gnorm_scale, bool skip_zeros) -> ()",
+    "(str optimizer_name, Tensor(a0!) g, Tensor(a1!) p, Tensor(a2!) state1, Tensor(a3!)? state2, float beta1, float beta2, float beta3, float alpha, float eps, int step, float lr, Tensor(a4!) qmap1, Tensor(a5!)? qmap2, Tensor(a6!) absmax1, Tensor(a7!)? absmax2, float weight_decay, float gnorm_scale, bool skip_zeros=False) -> ()",
 )
 
 
@@ -417,8 +417,8 @@ def _(
     qmap2: Optional[torch.Tensor],
     absmax1: torch.Tensor,
     absmax2: Optional[torch.Tensor],
-    weight_decay: float = 0.0,
-    gnorm_scale: float = 1.0,
+    weight_decay: float,
+    gnorm_scale: float,
     skip_zeros=False,
 ) -> None:
     torch._check(

bitsandbytes/backends/cuda/ops.py

@@ -686,8 +686,8 @@ def _optimizer_update_8bit_blockwise_impl(
     qmap2: Optional[torch.Tensor],
     absmax1: torch.Tensor,
     absmax2: Optional[torch.Tensor],
-    weight_decay: float = 0.0,
-    gnorm_scale: float = 1.0,
+    weight_decay: float,
+    gnorm_scale: float,
     skip_zeros=False,
 ) -> None:
     # torch._check(