Filter out redundant ops/allocations in no_grad mode

src/liger_kernel/ops/cross_entropy.py

@@ -414,6 +414,8 @@ def forward(
         Returns:
         tuple: A tuple with the compouted losses with respect to loss and z loss. The elements are tensors or None.
         """
+        input_requires_grad = _input.requires_grad
+
         loss, z_loss, _input = cross_entropy_forward(
             _input,
             target,
@@ -428,7 +430,8 @@ def forward(
         # TODO: investigation
         # If we don't detach the _input tensor, the memory will double
         # Not sure why but seems that there will be a time both grad and value exist but in different location
-        ctx.save_for_backward(_input.detach())
+        if input_requires_grad:
+            ctx.save_for_backward(_input.detach())
         ctx.return_z_loss = return_z_loss
 
         return loss, z_loss

src/liger_kernel/ops/fused_linear_cross_entropy.py

@@ -31,6 +31,8 @@ def fused_linear_cross_entropy_forward(
     assert isinstance(return_z_loss, bool), f"return_z_loss must be True or False. Got: {return_z_loss}"
     device = _input.device
 
+    input_requires_grad = _input.requires_grad
+
     # inputs have shape: BT x H
     # materialized activations will have shape: BT x V
     # the increase in memory = BT x V
@@ -49,12 +51,13 @@ def fused_linear_cross_entropy_forward(
     grad_input = torch.zeros_like(_input, device=device)
 
     # we use fp32 for loss and gradients accumulator
-    if accum_dtype is None:
-        grad_weight = torch.zeros_like(weight, device=device) if weight.requires_grad else None
-        grad_bias = torch.zeros_like(bias, device=device) if bias is not None else None
-    else:
-        grad_weight = torch.zeros_like(weight, dtype=accum_dtype, device=device) if weight.requires_grad else None
-        grad_bias = torch.zeros_like(bias, dtype=accum_dtype, device=device) if bias is not None else None
+    if input_requires_grad:
+        if accum_dtype is None:
+            grad_weight = torch.zeros_like(weight, device=device) if weight.requires_grad else None
+            grad_bias = torch.zeros_like(bias, device=device) if bias is not None else None
+        else:
+            grad_weight = torch.zeros_like(weight, dtype=accum_dtype, device=device) if weight.requires_grad else None
+            grad_bias = torch.zeros_like(bias, dtype=accum_dtype, device=device) if bias is not None else None
 
     loss_1d = torch.zeros(BT, dtype=torch.float32, device=device)
     z_loss_1d = torch.zeros(BT, dtype=_input.dtype, device=_input.device) if return_z_loss else None
@@ -150,7 +153,7 @@ def fused_linear_cross_entropy_forward(
             RETURN_Z_LOSS=return_z_loss,
             HAS_WEIGHT=True if ce_weight is not None else False,
             HAS_SOFTCAPPING=True if softcap is not None else False,
-            HAS_GRADIENTS=_input.requires_grad,
+            HAS_GRADIENTS=input_requires_grad,
             BLOCK_SIZE=BLOCK_SIZE,
             num_warps=32 if not is_hip() else 16,
         )
@@ -172,12 +175,13 @@ def fused_linear_cross_entropy_forward(
             scaling_factors_expanded = scaling_factors.unsqueeze(-1)  # chunk_size x 1
             grad_logits_chunk = grad_logits_chunk * scaling_factors_expanded
 
-        grad_input[start_idx:end_idx] = grad_logits_chunk @ weight
+        if input_requires_grad:
+            grad_input[start_idx:end_idx] = grad_logits_chunk @ weight
 
-        if grad_weight is not None and _input.requires_grad:
+        if grad_weight is not None and input_requires_grad:
             grad_weight += torch.mm(grad_logits_chunk.t(), _input_chunk).float()
 
-        if bias is not None and _input.requires_grad:
+        if bias is not None and input_requires_grad:
             torch.add(
                 input=grad_bias,
                 other=grad_logits_chunk.sum(dim=0),