Temporarily enforce eager mode for GPTQ models (#2154)

2023-12-17 01:51:12 -08:00
parent 26c52a5ea6
commit 3a765bd5e1
1 changed files with 5 additions and 0 deletions
--- a/vllm/config.py
+++ b/vllm/config.py
@@ -185,6 +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:
+            # Related issue: https://github.com/vllm-project/vllm/issues/2147
+            logger.warning("GPTQ does not support CUDA graph yet. Disabling "
+                           "CUDA graph.")
+            self.enforce_eager = True

    def verify_with_parallel_config(
        self,