Disable CUDA graph for SqueezeLLM (#2161)

vllm/config.py

@@ -185,10 +185,11 @@ class ModelConfig:
             self.max_context_len_to_capture = self.max_model_len
         self.max_context_len_to_capture = min(self.max_context_len_to_capture,
                                               self.max_model_len)
-        if self.quantization == "gptq" and not self.enforce_eager:
+        if (self.quantization in ["gptq", "squeezellm"]
+                and not self.enforce_eager):
             # Related issue: https://github.com/vllm-project/vllm/issues/2147
-            logger.warning("GPTQ does not support CUDA graph yet. Disabling "
-                           "CUDA graph.")
+            logger.warning(f"{self.quantization} does not support CUDA graph "
+                           "yet. Disabling CUDA graph.")
             self.enforce_eager = True
 
     def verify_with_parallel_config(