diff --git a/tests/integration/defs/accuracy/test_llm_api_pytorch.py b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
index 16c8184c6ee..0cc34f61e85 100644
--- a/tests/integration/defs/accuracy/test_llm_api_pytorch.py
+++ b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
@@ -263,9 +263,12 @@ def test_eagle3(self, overlap_scheduler, eagle3_one_model):
 
     @skip_pre_hopper
     def test_ngram(self):
+        max_bs = 16
+
         pytorch_config = dict(
             disable_overlap_scheduler=True,
-            cuda_graph_config=CudaGraphConfig(batch_sizes=[1]),
+            cuda_graph_config=CudaGraphConfig(
+                batch_sizes=[i for i in range(1, max_bs + 1)]),
         )
 
         kv_cache_config = KvCacheConfig(enable_block_reuse=False,
@@ -283,9 +286,7 @@ def test_ngram(self):
                  **pytorch_config,
                  kv_cache_config=kv_cache_config,
                  speculative_config=spec_config,
-                 max_batch_size=16) as llm:
-            task = MMLU(self.MODEL_NAME)
-            task.evaluate(llm)
+                 max_batch_size=max_bs) as llm:
             task = GSM8K(self.MODEL_NAME)
             task.evaluate(llm)
 
@@ -592,7 +593,7 @@ def test_fp8_eagle3_tp8(self, eagle3_one_model):
                                           speculative_model_dir=eagle_model_dir,
                                           eagle3_one_model=eagle3_one_model)
         pytorch_config = dict(
-            disable_overlap_scheduler=True,
+            disable_overlap_scheduler=not eagle3_one_model,
             cuda_graph_config=CudaGraphConfig(max_batch_size=1))
         with LLM(model_path,
                  max_batch_size=16,
@@ -1210,6 +1211,25 @@ def test_bfloat16(self, mtp_nextn, attention_dp, cuda_graph,
             task = GSM8K(self.MODEL_NAME)
             task.evaluate(llm)
 
+    @pytest.mark.skip_less_device_memory(60000)
+    def test_bfloat16_2_model_mtp(self):
+        kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.75)
+        pytorch_config = dict(
+            disable_overlap_scheduler=True,
+            cuda_graph_config=CudaGraphConfig(),
+        )
+        mtp_config = MTPDecodingConfig(num_nextn_predict_layers=3,
+                                       mtp_eagle_one_model=False,
+                                       speculative_model_dir=self.MODEL_PATH)
+        with LLM(self.MODEL_PATH,
+                 kv_cache_config=kv_cache_config,
+                 enable_chunked_prefill=False,
+                 max_num_tokens=8192,
+                 **pytorch_config,
+                 speculative_config=mtp_config) as llm:
+            task = GSM8K(self.MODEL_NAME)
+            task.evaluate(llm)
+
     @pytest.mark.skip_less_device(4)
     @parametrize_with_ids("torch_compile", [False, True])
     @parametrize_with_ids("attention_dp,cuda_graph,overlap_scheduler",
@@ -1854,14 +1874,19 @@ def test_chunked_prefill(self, quant_dtype, kv_cache_reuse, fp8kv,
 
     @parametrize_with_ids("mtp_nextn",
                           [0, pytest.param(2, marks=skip_pre_hopper)])
+    @parametrize_with_ids("use_one_model", [False, True])
     @pytest.mark.parametrize("backend", ["xgrammar", "llguidance"])
-    def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker):
+    def test_guided_decoding(self, backend: str, mtp_nextn: int,
+                             use_one_model: bool, mocker):
         mocker.patch.dict(os.environ, {"TRTLLM_XGUIDANCE_LENIENT": "1"})
         kv_cache_config = KvCacheConfig(free_gpu_memory_fraction=0.75)
         cuda_graph_config = CudaGraphConfig(enable_padding=True)
         mtp_config = None
         if mtp_nextn > 0:
-            mtp_config = MTPDecodingConfig(num_nextn_predict_layers=mtp_nextn)
+            mtp_config = MTPDecodingConfig(
+                num_nextn_predict_layers=mtp_nextn,
+                mtp_eagle_one_model=use_one_model,
+                speculative_model_dir=self.MODEL_PATH)
         llm = LLM(self.MODEL_PATH,
                   guided_decoding_backend=backend,
                   kv_cache_config=kv_cache_config,
@@ -3036,31 +3061,6 @@ def test_nvfp4(
             task = GSM8K(self.MODEL_NAME)
             task.evaluate(llm)
 
-    def test_eagle3(self):
-        pytorch_config = dict(
-            disable_overlap_scheduler=False,
-            cuda_graph_config=CudaGraphConfig(batch_sizes=[1, 2, 3, 4, 8]),
-        )
-        kv_cache_config = KvCacheConfig(enable_block_reuse=False)
-
-        eagle_model_dir = f"{llm_models_root()}/Qwen3/Qwen3-30B-eagle3"
-        target_model_dir = f"{llm_models_root()}/Qwen3/Qwen3-30B-A3B"
-
-        draft_len = 1
-        spec_config = EagleDecodingConfig(max_draft_len=draft_len,
-                                          speculative_model_dir=eagle_model_dir,
-                                          eagle3_one_model=True)
-
-        llm = LLM(model=target_model_dir,
-                  **pytorch_config,
-                  kv_cache_config=kv_cache_config,
-                  speculative_config=spec_config,
-                  max_seq_len=8192)
-
-        with llm:
-            task = GSM8K(self.MODEL_NAME)
-            task.evaluate(llm)
-
     @pytest.mark.parametrize("moe_backend", ["CUTLASS", "TRITON", "TRTLLM"])
     @pytest.mark.parametrize(
         "tp_size,pp_size,ep_size,attention_dp,cuda_graph,overlap_scheduler", [
diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt
index 460f1da4930..71f1610282c 100644
--- a/tests/integration/test_lists/qa/llm_function_core.txt
+++ b/tests/integration/test_lists/qa/llm_function_core.txt
@@ -458,6 +458,7 @@ accuracy/test_llm_api_pytorch.py::TestMixtral8x7B::test_fp8_tp2
 accuracy/test_llm_api_pytorch.py::TestMixtral8x7B::test_nvfp4_tp2
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True]
+accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]
@@ -467,10 +468,12 @@ accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_online
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_online_eplb[mtp_nextn=2]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus_online_eplb[fp8kv=False]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus_online_eplb[fp8kv=True]
-accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]
-accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]
-accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0]
-accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2]
+accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-use_one_model=True-mtp_nextn=0]
+accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-use_one_model=True-mtp_nextn=2]
+accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-use_one_model=True-mtp_nextn=0]
+accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-use_one_model=True-mtp_nextn=2]
+accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-use_one_model=False-mtp_nextn=2]
+accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-use_one_model=False-mtp_nextn=2]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=0]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=2]
 accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[llguidance-mtp_nextn=0]
diff --git a/tests/integration/test_lists/test-db/l0_b200.yml b/tests/integration/test_lists/test-db/l0_b200.yml
index 114717be909..f6810956fb2 100644
--- a/tests/integration/test_lists/test-db/l0_b200.yml
+++ b/tests/integration/test_lists/test-db/l0_b200.yml
@@ -22,6 +22,7 @@ l0_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False] ISOLATION
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
@@ -33,7 +34,8 @@ l0_b200:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=False-fp8kv=False-overlap_scheduler=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]
-  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-use_one_model=False-mtp_nextn=2]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-use_one_model=True-mtp_nextn=2]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_dummy_load_format
   - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[True-True-cutlass-auto]
   - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[True-True-trtllm-auto]
diff --git a/tests/integration/test_lists/test-db/l0_h100.yml b/tests/integration/test_lists/test-db/l0_h100.yml
index 3b7d94d38c3..f5f4a69c84d 100644
--- a/tests/integration/test_lists/test-db/l0_h100.yml
+++ b/tests/integration/test_lists/test-db/l0_h100.yml
@@ -54,6 +54,7 @@ l0_h100:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=vanilla-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False]
+  - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=fp8-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=False-fp8kv=False-overlap_scheduler=True]
@@ -242,7 +243,7 @@ l0_h100:
   - accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]
   - 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.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-use_one_model=True-mtp_nextn=0]
   - 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)