[CI][BugFix] Adapt test_token_penalty to Paddle uniform float16 behavior change (#8031)

Author: EmmonsCurse

tests/cov_pytest.ini

@@ -14,4 +14,3 @@ addopts =
     --ignore=tests/graph_optimization/test_cuda_graph_dynamic_subgraph.py
     --ignore=tests/e2e/test_ernie_03b_pd_decode_unified_attention.py
     --ignore=tests/e2e/test_ernie_03b_pd_router_v1_ipc.py
-    --ignore=tests/operators/test_token_penalty.py

tests/operators/test_token_penalty.py

@@ -44,8 +44,10 @@ def test_token_penalty_once(self):
             for orig, penal in zip(original_values.numpy(), penalized_values.numpy()):
                 if orig < 0:
                     self.assertLess(penal, orig, "负值应该乘以惩罚因子")
-                else:
+                elif orig > 0:
                     self.assertLess(penal, orig, "正值应该除以惩罚因子")
+                else:
+                    self.assertEqual(penal, orig, "零值不受惩罚影响")
 
     def test_compare_with_naive_implementation(self):
         res = get_token_penalty_once(self.pre_ids, self.logits, self.penalty_scores)