[NPU] Fix evalscope accuracy parsing and add glm5_1 aime26 request timeout (#36459)

test/registered/npu/accuracy/glm5_1/test_npu_glm5_1_w4a8_1p1d_32p_in64k_out1k_50ms_aime26.py

@@ -159,7 +159,7 @@ GLM_5_1_PD_SEP_MODEL_CONFIG = {
     "decode_args": GLM_5_1_PD_SEP_DECODE_ARGS,
     "prefill_envs": GLM_5_1_PD_SEP_PREFILL_ENVS,
     "decode_envs": GLM_5_1_PD_SEP_DECODE_ENVS,
-    "router_args": ["--policy", "round_robin"],
+    "router_args": ["--policy", "round_robin", "--request-timeout-secs", 7200],
     "router_envs": {},
 }
 

test/registered/npu/accuracy/qwen3_next_80b_a3b_instruct/test_npu_qwen3_next_80b_w8a8_2p_in6k_out1k5_bs16_aime25.py

@@ -112,7 +112,7 @@ class TestQwen3Next80BA3B_aime25(TestNpuAccuracyTestCaseBase):
     max_concurrency = 16
 
     def test_aime25(self):
-        self.run_accuracy_multiple(n_runs=3)
+        self.run_accuracy()
 
 
 if __name__ == "__main__":

test/registered/npu/performance/kimi_k2_6/test_npu_kimi_k2_6_w4a8_16p_in64k_out1k_100ms.py

@@ -44,7 +44,7 @@ OTHER_ARGS = [
     "--nnodes",
     2,
     "--mem-fraction-static",
-    0.662,
+    0.68,
     "--max-running-requests",
     32,
     "--chunked-prefill-size",