fix bugs of GPTOSS CIs and unwaive them

Signed-off-by: bhsueh <[email protected]>

Author: byshiue

tests/integration/defs/accuracy/test_llm_api_pytorch.py

@@ -2704,8 +2704,6 @@ class TestGPTOSS(LlmapiAccuracyTestHarness):
     extra_evaluator_kwargs = {
         "fewshot_as_multiturn": True,
         "apply_chat_template": True,
-        "scores_filter": "exact_match,flexible-extract",
-        "MAX_OUTPUT_LEN": 8192
     }
 
     MODEL_PATH = f"{llm_models_root()}/gpt_oss/gpt-oss-120b"
@@ -2736,7 +2734,8 @@ def test_w4_1gpu(self, moe_backend, cuda_graph, overlap_scheduler, mocker):
 
         with llm:
             model_name = "GPT-OSS/MXFP4"
-            mocker.patch.object(GSM8K, "MAX_OUTPUT_LEN": 8192)
+            mocker.patch.object(GSM8K, "MAX_OUTPUT_LEN", 8192)
+            mocker.patch.dict(GSM8K.EVALUATE_KWARGS, {"scores_filter": "exact_match,flexible-extract"})
             task = GSM8K(model_name)
             task.evaluate(llm,
                           extra_evaluator_kwargs=self.extra_evaluator_kwargs)
@@ -2776,7 +2775,8 @@ def test_w4_4gpus(self, moe_backend, tp_size, pp_size, ep_size,
         with llm:
             model_name = "GPT-OSS/MXFP4"
             task = GSM8K(model_name)
-            mocker.patch.object(GSM8K, "MAX_OUTPUT_LEN": 8192)
+            mocker.patch.object(GSM8K, "MAX_OUTPUT_LEN", 8192)
+            mocker.patch.dict(GSM8K.EVALUATE_KWARGS, {"scores_filter": "exact_match,flexible-extract"})
             task.evaluate(llm,
                           extra_evaluator_kwargs=self.extra_evaluator_kwargs)
 
@@ -2807,7 +2807,8 @@ def test_w4a16(self, tp_size, pp_size, ep_size, attention_dp, cuda_graph,
         with llm:
             model_name = "GPT-OSS/BF16"
             task = GSM8K(model_name)
-            mocker.patch.object(GSM8K, "MAX_OUTPUT_LEN": 8192)
+            mocker.patch.object(GSM8K, "MAX_OUTPUT_LEN", 8192)
+            mocker.patch.dict(GSM8K.EVALUATE_KWARGS, {"scores_filter": "exact_match,flexible-extract"})
             task.evaluate(llm,
                           extra_evaluator_kwargs=self.extra_evaluator_kwargs)
 

tests/integration/test_lists/waives.txt

@@ -325,10 +325,6 @@ accuracy/test_cli_flow.py::TestPhi4MiniInstruct::test_tp2 SKIP (https://nvbugs/5
 accuracy/test_cli_flow.py::TestLongAlpaca7B::test_auto_dtype SKIP (https://nvbugs/5481075)
 accuracy/test_llm_api.py::TestPhi4MiniInstruct::test_fp8 SKIP (https://nvbugs/5465143)
 accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput] SKIP (https://nvbugs/5471106)
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[True-True-cutlass] SKIP (https://nvbugs/5481080)
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-cutlass] SKIP (https://nvbugs/5481080)
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass] SKIP (https://nvbugs/5481080)
-accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass] SKIP (https://nvbugs/5481080)
 accuracy/test_llm_api_pytorch.py::TestEXAONE4::test_auto_dtype SKIP (https://nvbugs/5481090)
 test_e2e.py::test_ptp_quickstart_advanced_8gpus_chunked_prefill_sq_22k[Llama-4-Maverick-17B-128E-Instruct-FP8-llama4-models/nvidia/Llama-4-Maverick-17B-128E-Instruct-FP8-False] SKIP (https://nvbugs/5481094)
 test_e2e.py::test_ptp_quickstart_advanced_8gpus_chunked_prefill_sq_22k[Llama-4-Maverick-17B-128E-Instruct-FP8-llama4-models/nvidia/Llama-4-Maverick-17B-128E-Instruct-FP8-True] SKIP (https://nvbugs/5481094)