[https://nvbugs/5837281][fix] Fix trtllm-serve guided decoding test (#11101)

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

Author: syuoni

tests/integration/defs/test_e2e.py

@@ -1736,8 +1736,10 @@ def test_openai_mmencoder_example(llm_root, llm_venv):
          str(test_root / "_test_openai_mmencoder.py")])
 
 
-@pytest.mark.parametrize(
-    "model_name", ["meta-llama/Llama-3.1-8B-Instruct", "openai/gpt-oss-120b"])
+@pytest.mark.parametrize("model_name", [
+    "meta-llama/Llama-3.1-8B-Instruct",
+    pytest.param("openai/gpt-oss-120b", marks=skip_pre_hopper)
+])
 def test_openai_chat_guided_decoding(llm_root, llm_venv, model_name: str):
     test_root = unittest_path() / "llmapi" / "apps"
     llm_venv.run_cmd([

tests/integration/test_lists/waives.txt

@@ -344,7 +344,6 @@ examples/test_llama.py::test_llama_3_x_with_bf16_lora_torch[llama-3.2-1b-instruc
 accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_bf16 SKIP (https://nvbugs/5838184)
 cpp/test_multi_gpu.py::test_cache_transceiver[8proc-mooncake_kvcache-90] SKIP (https://nvbugs/5838199)
 accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[dp4-cutlass-auto] SKIP (https://nvbugs/5838211)
-test_e2e.py::test_openai_chat_guided_decoding[openai/gpt-oss-120b] SKIP (https://nvbugs/5836594)
 test_e2e.py::test_ptp_quickstart_advanced_ngram[Llama-3.1-8B-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct] SKIP (https://nvbugs/5843112)
 accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput_mtp] SKIP (https://nvbugs/5839028)
 unittest/kv_cache_manager_v2_tests/test_kv_cache_manager_v2.py::TestNoBatching::test_naive_0 SKIP (https://nvbugs/5841954)

tests/unittest/llmapi/apps/_test_openai_chat_guided_decoding.py

@@ -9,11 +9,16 @@
 import openai
 import pytest
 import yaml
+from utils.llm_data import llm_datasets_root
 
 from ..test_llm import get_model_path
 from .openai_server import RemoteOpenAIServer
 
 pytestmark = pytest.mark.threadleak(enabled=False)
+os.environ['TIKTOKEN_RS_CACHE_DIR'] = os.path.join(llm_datasets_root(),
+                                                   'tiktoken_vocab')
+os.environ['TIKTOKEN_ENCODINGS_BASE'] = os.path.join(llm_datasets_root(),
+                                                     'tiktoken_vocab')
 
 
 @pytest.fixture(