Update docstring in examples

Author: FKKimura

example/.gitignore

@@ -4,3 +4,4 @@ tinyllama*
 TinyLlama*/
 quantized_model_saveload/
 rotated_model_*
+results

example/vllm_inference/example_autobit_vllm_inference.py

@@ -10,6 +10,18 @@
 Requirements:
   pip install vllm
 
+Note:
+  vLLM runs a DeepGEMM (FP8) kernel warmup at engine startup even for
+  non-FP8 quantization such as mixed-precision GPTQ. If ``deep_gemm`` is not
+  installed this fails with ``RuntimeError: DeepGEMM backend is not available
+  or outdated``. OneComp-quantized models do not need DeepGEMM, so disable the
+  FP8 path before running this script::
+
+      export VLLM_USE_DEEP_GEMM=0
+      export VLLM_DEEP_GEMM_WARMUP=skip
+
+  See docs/user-guide/vllm-inference.md (Troubleshooting) for details.
+
 Copyright 2025-2026 Fujitsu Ltd.
 
 Author: Keiji Kimura

example/vllm_inference/example_gptq_vllm_inference.py

@@ -11,6 +11,18 @@
 Requirements:
   pip install vllm
 
+Note:
+  vLLM runs a DeepGEMM (FP8) kernel warmup at engine startup even for
+  non-FP8 quantization such as GPTQ. If ``deep_gemm`` is not installed this
+  fails with ``RuntimeError: DeepGEMM backend is not available or outdated``.
+  OneComp-quantized models do not need DeepGEMM, so disable the FP8 path
+  before running this script::
+
+      export VLLM_USE_DEEP_GEMM=0
+      export VLLM_DEEP_GEMM_WARMUP=skip
+
+  See docs/user-guide/vllm-inference.md (Troubleshooting) for details.
+
 Copyright 2025-2026 Fujitsu Ltd.
 
 Author: Keiji Kimura

example/vllm_inference/example_jointq_vllm_inference.py

@@ -22,6 +22,18 @@
 Requirements:
   pip install vllm
 
+Note:
+  vLLM runs a DeepGEMM (FP8) kernel warmup at engine startup even for
+  non-FP8 quantization. If ``deep_gemm`` is not installed this fails with
+  ``RuntimeError: DeepGEMM backend is not available or outdated``.
+  OneComp-quantized models do not need DeepGEMM, so disable the FP8 path
+  before running this script::
+
+      export VLLM_USE_DEEP_GEMM=0
+      export VLLM_DEEP_GEMM_WARMUP=skip
+
+  See docs/user-guide/vllm-inference.md (Troubleshooting) for details.
+
 Copyright 2025-2026 Fujitsu Ltd.
 
 Author: Keiji Kimura