Fix stale CUDA graph benchmark and docs refs (#28041)

benchmark/kernels/fused_moe_triton/benchmark_torch_compile_fused_moe.py

@@ -7,10 +7,10 @@ from torch.nn import functional as F
 from transformers import AutoConfig
 
 from sglang.benchmark.bench_utils import run_bench
+from sglang.srt.compilation.torch_compile_decoration import set_torch_compile_config
 from sglang.srt.layers.moe.moe_runner.triton_utils.fused_moe import (
     fused_moe as fused_moe_triton,
 )
-from sglang.srt.model_executor.cuda_graph_runner import set_torch_compile_config
 
 
 def get_model_config(model_name: str, tp_size: int):

docs_new/docs/advanced_features/breakable_cuda_graph.mdx

@@ -33,7 +33,7 @@ This mode is intended for debugging only — it eliminates the performance benef
 For production use, you can mark specific functions as "non-graphable" using the `@eager_on_graph` decorator. During CUDA graph capture, these functions run eagerly between captured graph segments. Outside of capture, they behave normally.
 
 ```python
-from sglang.srt.model_executor.breakable_cuda_graph.breakable_cuda_graph import eager_on_graph
+from sglang.srt.model_executor.runner_backend_utils.breakable_cuda_graph import eager_on_graph
 
 @eager_on_graph(enable=True)
 def my_dynamic_op(x):
@@ -44,7 +44,7 @@ def my_dynamic_op(x):
 You can also insert a bare graph break (no computation) using the `break_graph()` helper:
 
 ```python
-from sglang.srt.model_executor.breakable_cuda_graph.breakable_cuda_graph import break_graph
+from sglang.srt.model_executor.runner_backend_utils.breakable_cuda_graph import break_graph
 
 def forward(self, x):
     x = self.layer1(x)
@@ -169,16 +169,16 @@ For typical use cases with a small number of graph breaks, the overhead is negli
   </thead>
   <tbody>
     <tr>
-      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/breakable_cuda_graph/breakable_cuda_graph.py</code></td>
+      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/runner_backend_utils/breakable_cuda_graph/breakable_cuda_graph.py</code></td>
       <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Core implementation: <code>eager_on_graph</code>, <code>BreakableCUDAGraph</code>, <code>BreakableCUDAGraphCapture</code></td>
     </tr>
     <tr>
-      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/breakable_cuda_graph/cuda_utils.py</code></td>
+      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/runner_backend_utils/breakable_cuda_graph/cuda_utils.py</code></td>
       <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>CUDA runtime binding utilities</td>
     </tr>
     <tr>
-      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/cuda_graph_runner.py</code></td>
+      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/runner_backend/breakable_cuda_graph_backend.py</code></td>
-      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Integration with the main CUDA graph runner</td>
+      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Integration with CUDA graph runner backends</td>
     </tr>
     <tr>
       <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/server_args.py</code></td>

docs_new/docs/advanced_features/piecewise_cuda_graph.mdx

@@ -264,8 +264,8 @@ Use `--enforce-piecewise-cuda-graph` to skip all auto-disable checks (for testin
   </thead>
   <tbody>
     <tr>
-      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/piecewise_cuda_graph_runner.py</code></td>
+      <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/model_executor/runner_backend/tc_piecewise_cuda_graph_backend.py</code></td>
-      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Main runner: init, capture, replay</td>
+      <td style={{padding: "9px 12px", backgroundColor: "rgba(255,255,255,0.05)"}}>Backend implementation: compile, capture, replay</td>
     </tr>
     <tr>
       <td style={{padding: "9px 12px", fontWeight: 500, backgroundColor: "rgba(255,255,255,0.02)"}}><code>python/sglang/srt/compilation/compile.py</code></td>

python/sglang/srt/model_executor/runner_backend_utils/breakable_cuda_graph/__init__.py

@@ -3,16 +3,16 @@
 Public API (also reachable via the deeper module paths):
   - BreakableCUDAGraph, BreakableCUDAGraphCapture — capture/replay
   - eager_on_graph — decorator that marks a callable as a graph break
+  - break_graph — helper that inserts a bare graph break
   - enable_breakable_cuda_graph — context that flips the Breakable runtime flag
   - is_in_breakable_cuda_graph — runtime flag getter
 
-The legacy model_executor/breakable_cuda_graph/ package is a
-backwards-compat shim that re-exports from here.
 """
 
 from sglang.srt.model_executor.runner_backend_utils.breakable_cuda_graph.breakable_cuda_graph import (  # noqa: F401
     BreakableCUDAGraph,
     BreakableCUDAGraphCapture,
+    break_graph,
     eager_on_graph,
 )
 from sglang.srt.model_executor.runner_backend_utils.breakable_cuda_graph.context import (  # noqa: F401