[torch.compile] remove compilation_context and simplify code (#10838)

Signed-off-by: youkaichao <youkaichao@gmail.com>

tests/models/decoder_only/language/test_jamba.py

@@ -1,8 +1,8 @@
 import pytest
 
 from tests.utils import multi_gpu_test
+from vllm.config import VllmConfig
 from vllm.sampling_params import SamplingParams
-from vllm.worker.model_runner import _get_graph_batch_size
 
 from ...utils import check_outputs_equal
 
@@ -189,7 +189,8 @@ def test_mamba_cache_cg_padding(
     # This test is for verifying that mamba cache is padded to CG captured
     # batch size. If it's not, a torch RuntimeError will be raised because
     # tensor dimensions aren't compatible
-    while len(example_prompts) == _get_graph_batch_size(len(example_prompts)):
+    while len(example_prompts) == VllmConfig.get_graph_batch_size(
+            len(example_prompts)):
         example_prompts.append(example_prompts[0])
 
     try: