FIX TST Small regression in BNB LoRA output (#2238)

Our regression tests reveal that the 8bit LoRA BNB regression test is
failing. To reproduce, run:

pytest tests/regression/test_regression.py -s --regression -k
test_lora_8bit

The regression was introduced in #2122. We didn't notice this earlier
because of other failing tests in the nightly CI.

The cause of the error is subtle. In the original code, we would
calculate the LoRA output, convert the dtype if necessary, then add it
to the base output. After the mentioned PR, we calculate the LoRA
output, add it to the base output, then convert the dtype if necessary.
The difference is very small on a per layer basis, but it can accumulate
over the layers, leading to a significant difference in outputs, as
witnessed by the regression test.

This PR rolls back this specific part of the PR (both for 8bit and 4bit)
while leaving the main change of that PR intact.

src/peft/tuners/lora/bnb.py

@@ -248,15 +248,15 @@ def forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
                             x = x.to(compute_dtype)
 
                     if not self.use_dora[active_adapter]:
-                        result = result + lora_B(lora_A(dropout(x))) * scaling
+                        output = lora_B(lora_A(dropout(x))) * scaling
                     else:
                         if isinstance(dropout, torch.nn.Identity) or not self.training:
                             base_result = result
                         else:
                             x = dropout(x)
                             base_result = None
 
-                        result = result + self.lora_magnitude_vector[active_adapter](
+                        output = self.lora_magnitude_vector[active_adapter](
                             x,
                             lora_A=lora_A,
                             lora_B=lora_B,
@@ -265,7 +265,8 @@ def forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
                             base_result=base_result,
                         )
                     if requires_conversion:
-                        result = result.to(expected_dtype)
+                        output = output.to(expected_dtype)
+                    result = result + output
 
             return result
 
@@ -515,15 +516,15 @@ def forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
                         x = x.to(lora_A.weight.dtype)
 
                     if not self.use_dora[active_adapter]:
-                        result = result + lora_B(lora_A(dropout(x))) * scaling
+                        output = lora_B(lora_A(dropout(x))) * scaling
                     else:
                         if isinstance(dropout, torch.nn.Identity) or not self.training:
                             base_result = result
                         else:
                             x = dropout(x)
                             base_result = None
 
-                        result = result + self.lora_magnitude_vector[active_adapter](
+                        output = self.lora_magnitude_vector[active_adapter](
                             x,
                             lora_A=lora_A,
                             lora_B=lora_B,
@@ -532,7 +533,8 @@ def forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
                             base_result=base_result,
                         )
                     if requires_conversion:
-                        result = result.to(expected_dtype)
+                        output = output.to(expected_dtype)
+                    result = result + output
 
             return result