[nvbug/5410296][fix] Fix OOM in Llama 4 disagg-serve tests (#6439)

Signed-off-by: Bo Deng <[email protected]>

Author: bo-nv

tests/integration/defs/accuracy/test_disaggregated_serving.py

@@ -413,6 +413,9 @@ def test_auto_dtype(self, overlap_scheduler):
         gen_server_config = {"disable_overlap_scheduler": overlap_scheduler}
         ctx_server_config["cache_transceiver_config"] = {"backend": "default"}
         gen_server_config["cache_transceiver_config"] = {"backend": "default"}
+        # Keep this low to avoid warmup OOM in CI
+        ctx_server_config["max_seq_len"] = 8192
+        gen_server_config["max_seq_len"] = 8192
         disaggregated_server_config = {
             "hostname": "localhost",
             "port": 8000,

tests/integration/defs/disaggregated/test_disaggregated.py

@@ -745,7 +745,6 @@ def test_disaggregated_deepseek_v3_lite_fp8_ucx(disaggregated_test_root,
                            cwd=llm_venv.get_working_directory())
 
 
-@skip_no_hopper
 @skip_arm
 @pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'],
                          indirect=True)

tests/integration/test_lists/waives.txt

@@ -419,8 +419,6 @@ accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_ngram SKIP (http
 test_e2e.py::test_openai_multi_chat_example SKIP (https://nvbugs/5409416)
 test_e2e.py::test_ptp_quickstart_multimodal[llava-v1.6-mistral-7b-llava-v1.6-mistral-7b-hf-image-False] SKIP (https://nvbugs/5409417)
 test_e2e.py::test_ptp_star_attention_example[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B] SKIP (https://nvbugs/5409420)
-accuracy/test_disaggregated_serving.py::TestLlama4ScoutInstruct::test_auto_dtype[False] SKIP (https://nvbugs/5410296)
-accuracy/test_disaggregated_serving.py::TestLlama4ScoutInstruct::test_auto_dtype[True] SKIP (https://nvbugs/5410296)
 llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_mtp SKIP (https://nvbugs/5410399)
 unittest/trt/attention/test_gpt_attention.py -k "partition0" SKIP (https://nvbugs/5412456)
 unittest/trt/attention/test_gpt_attention.py -k "partition1" SKIP (https://nvbugs/5412456)