feat: Add vLLM counter metrics access through Triton (#7493) (#7546)

Co-authored-by: Yingge He <[email protected]>

Author: mc-nv

build.py

@@ -1806,6 +1806,10 @@ def backend_clone(
         os.path.join(build_dir, be, "src", "model.py"),
         backend_dir,
     )
+    clone_script.cpdir(
+        os.path.join(build_dir, be, "src", "utils"),
+        backend_dir,
+    )
 
     clone_script.comment()
     clone_script.comment(f"end '{be}' backend")

docs/user_guide/metrics.md

@@ -378,3 +378,9 @@ Further documentation can be found in the `TRITONSERVER_MetricFamily*` and
 The TRT-LLM backend uses the custom metrics API to track and expose specific metrics about
 LLMs, KV Cache, and Inflight Batching to Triton:
 https://github.com/triton-inference-server/tensorrtllm_backend?tab=readme-ov-file#triton-metrics
+
+### vLLM Backend Metrics
+
+The vLLM backend uses the custom metrics API to track and expose specific metrics about
+LLMs to Triton:
+https://github.com/triton-inference-server/vllm_backend?tab=readme-ov-file#triton-metrics