Add two MTP disaggregated test (#4546)

Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>

tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_two_mtp.yaml

@@ -0,0 +1,24 @@
+hostname: localhost
+port: 8000
+model: DeepSeek-V3-Lite/fp8
+free_gpu_memory_fraction: 0.1
+backend: "pytorch"
+use_cuda_graph: False
+disable_overlap_scheduler: True
+speculative_config:
+  decoding_type: MTP
+  num_nextn_predict_layers: 2
+context_servers:
+  num_instances: 1
+  tensor_parallel_size: 1
+  pipeline_parallel_size: 1
+  enable_attention_dp: true
+  urls:
+      - "localhost:8001"
+generation_servers:
+  num_instances: 1
+  tensor_parallel_size: 1
+  pipeline_parallel_size: 1
+  enable_attention_dp: false
+  urls:
+      - "localhost:8002"

tests/integration/defs/disaggregated/test_disaggregated.py

@@ -106,6 +106,10 @@ def get_test_config(test_desc, example_dir, test_root):
          ),
         "deepseek_v3_lite_bf16_conditional":
         (2, f"{test_configs_root}/disagg_config_conditional_deepseek_v3.yaml"),
+        "deepseek_v3_lite_fp8_tp1_two_mtp":
+        (2,
+         f"{test_configs_root}/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_two_mtp.yaml"
+         ),
     }
 
     if test_desc not in config_map:
@@ -808,3 +812,25 @@ def test_disaggregated_deepseek_v3_lite_bf16_conditional(
                            "deepseek_v3_lite_bf16_conditional",
                            env=llm_venv._new_env,
                            cwd=llm_venv.get_working_directory())
+
+
+@skip_no_hopper
+@pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'],
+                         indirect=True)
+def test_disaggregated_deepseek_v3_lite_fp8_tp1_two_mtp(
+        disaggregated_test_root, disaggregated_example_root, llm_venv,
+        deepseek_v3_model_root):
+    src_dst_dict = {
+        deepseek_v3_model_root:
+        f"{llm_venv.get_working_directory()}/DeepSeek-V3-Lite/fp8",
+    }
+
+    for src, dst in src_dst_dict.items():
+        if not os.path.islink(dst):
+            os.makedirs(os.path.dirname(dst), exist_ok=True)
+            os.symlink(src, dst, target_is_directory=True)
+
+    run_disaggregated_test(disaggregated_example_root,
+                           "deepseek_v3_lite_fp8_tp1_two_mtp",
+                           env=llm_venv._new_env,
+                           cwd=llm_venv.get_working_directory())

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

@@ -62,6 +62,7 @@ l0_h100:
   - test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-instruct-hf-fp8-True-True]
   - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu[DeepSeek-V3-Lite-fp8]
   - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu_mtp[DeepSeek-V3-Lite-fp8]
+  - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_two_mtp[DeepSeek-V3-Lite-fp8]
   - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]
   - disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-False-TinyLlama-1.1B-Chat-v1.0]
   - disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-True-TinyLlama-1.1B-Chat-v1.0]