Fix/liger fused linear cross entropy function does not support reduction=none

src/liger_kernel/chunked_loss/cpo_loss.py

@@ -36,7 +36,7 @@ def preference_loss_fn(
         """
         logits = beta * (chosen_logps - rejected_logps)
         loss = (
-            - F.logsigmoid(logits) * (1 - label_smoothing)
+            -F.logsigmoid(logits) * (1 - label_smoothing)
             - F.logsigmoid(-logits) * label_smoothing
         ).sum() / (full_target.shape[0] // 2)
 

src/liger_kernel/chunked_loss/simpo_loss.py

@@ -42,7 +42,7 @@ def preference_loss_fn(
         """
         logits = beta * (chosen_logps - rejected_logps) - gamma
         loss = (
-            - F.logsigmoid(logits) * (1 - label_smoothing)
+            -F.logsigmoid(logits) * (1 - label_smoothing)
             - F.logsigmoid(-logits) * label_smoothing
         ).sum() / (full_target.shape[0] // 2)
 

src/liger_kernel/ops/fused_linear_cross_entropy.py

@@ -133,15 +133,18 @@ def fused_linear_cross_entropy_forward(
                 alpha=alpha,
             )
 
-    loss = torch.sum(loss_1d)
+    if reduction == "none":
+        loss = loss_1d
+    else:
+        loss = torch.sum(loss_1d)
     return loss, grad_input, grad_weight, grad_bias
 
 
 def fused_linear_cross_entropy_backward(
     grad_output, grad_input, grad_weight, grad_bias
 ):
     # If cross entropy is the last layer, grad_output is 1.0. Skip the mul to save time
-    if torch.ne(grad_output, torch.tensor(1.0, device=grad_output.device)):
+    if not torch.equal(grad_output, torch.tensor(1.0, device=grad_output.device)):
         # We use a Triton kernel instead of a PyTorch operation because modifying inputs in-place
         # for gradient storage and backward multiple times causes anomalies with PyTorch but not with Triton.
         BT, H = grad_input.shape

test/transformers/test_fused_linear_cross_entropy.py

@@ -114,6 +114,8 @@ def forward(self, x, y):
         ("mean", 1.0, torch.float32, 1e-5, 5e-4),
         ("sum", 1.0, torch.bfloat16, 5e-0, 5e1),
         ("sum", 1.0, torch.float32, 1e-3, 5e-2),
+        ("none", 1.0, torch.bfloat16, 5e-0, 5e1),
+        ("none", 1.0, torch.float32, 1e-3, 5e-2),
     ],
 )
 @pytest.mark.parametrize("bias", [True, False])
@@ -197,8 +199,8 @@ def test_correctness(
 
     assert_verbose_allclose(output1, output2, atol=atol, rtol=rtol)
 
-    output1.backward()
-    output2.backward()
+    output1.backward(gradient=torch.ones_like(output1))
+    output2.backward(gradient=torch.ones_like(output2))
 
     assert_verbose_allclose(_input1.grad, _input2.grad, atol=atol, rtol=rtol)