[https://nvbugs/5744427][fix] Fix accuracy test OOM (NVIDIA#10173)

Signed-off-by: Balaram Buddharaju <[email protected]>

Author: brb-nv

tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py

@@ -219,7 +219,7 @@ def test_auto_dtype(self):
 class TestGemma3_27BInstruct(LlmapiAccuracyTestHarness):
     MODEL_NAME = "google/gemma-3-27b-it"
     MODEL_PATH = f"{llm_models_root()}/gemma/gemma-3-27b-it/"
-    MAX_NUM_TOKENS = 25600
+    MAX_NUM_TOKENS = 12800
 
     sampling_params = SamplingParams(
         max_tokens=MAX_NUM_TOKENS, truncate_prompt_tokens=MMMU.MAX_INPUT_LEN, stop="<end_of_turn>"
@@ -229,7 +229,7 @@ class TestGemma3_27BInstruct(LlmapiAccuracyTestHarness):
     kv_cache_config = KvCacheConfig(
         enable_block_reuse=False,
         enable_partial_reuse=False,
-        free_gpu_memory_fraction=0.6,
+        free_gpu_memory_fraction=0.4,
     )
 
     def test_auto_dtype(self):

tests/integration/test_lists/test-db/l0_h100.yml

@@ -47,7 +47,6 @@ l0_h100:
   - accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_no_partial_reuse
   - accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_partial_reuse
   - accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_disable_overlap_scheduler
-  - accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_auto_dtype
   - accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=True]
   - accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[attn_backend=TRTLLM] TIMEOUT (90)
@@ -261,6 +260,7 @@ l0_h100:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=vanilla-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=none-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]
+  - accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_auto_dtype
   - accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized
   - accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_fp8_prequantized
   - accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_auto_dtype TIMEOUT (90)

tests/integration/test_lists/waives.txt

@@ -449,7 +449,6 @@ examples/test_phi.py::test_phi_fp8_with_bf16_lora[phi-2] SKIP (https://nvbugs/57
 examples/test_phi.py::test_llm_phi_1node_2gpus_summary[Phi-3.5-MoE-instruct-nb:1] SKIP (https://nvbugs/5744293)
 examples/test_phi.py::test_llm_phi_quantization_1gpu[phi-2-fp8-bfloat16] SKIP (https://nvbugs/5744293)
 disaggregated/test_disaggregated.py::test_disaggregated_trtllm_sampler[TinyLlama-1.1B-Chat-v1.0] SKIP (https://nvbugs/5741884)
-accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_auto_dtype SKIP (https://nvbugs/5744427)
 test_e2e.py::test_trtllm_bench_llmapi_launch[pytorch_backend-llama-v3-llama3-8b] SKIP (https://nvbugs/5744432)
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True] SKIP (https://nvbugs/5740087)
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True] SKIP (https://nvbugs/5740075)
@@ -492,3 +491,4 @@ examples/test_qwenvl.py::test_llm_qwenvl_single_gpu_summary[qwen-vl-chat] SKIP (
 examples/test_whisper.py::test_llm_whisper_general[large-v3-disable_gemm_plugin-enable_attention_plugin-int8-float16-nb:1-use_cpp_runtime] SKIP (https://nvbugs/5568052)
 accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3VL_MOE::test_auto_dtype SKIP (https://nvbugs/5588376)
 accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp2] SKIP (https://nvbugs/5756008)
+unittest/executor/test_base_worker.py::TestWorkerBase SKIP (https://nvbugs/5759698)