TiledFusedLogitsLoss bug fix (#7459)

bug fix - mixed up tuple and list.

Author: stas00

deepspeed/runtime/sequence_parallel/ulysses_sp.py

@@ -1012,7 +1012,7 @@ def forward(
             with torch.enable_grad():
                 args = (self, x_shard, y_shard)
                 if mask is not None:
-                    args.append(mask_shards[i])
+                    args += (mask_shards[i], )
                 output = fn(*args)
                 output_shards.append(output)
             torch.autograd.backward(output, incoming_grad)