enforce eager mode with bnb quantization temporarily (#6846)

2024-07-26 18:32:20 -07:00 · 2024-07-26 18:32:20 -07:00 · bb5494676f
commit bb5494676f
parent b5f49ee55b
1 changed files with 4 additions and 0 deletions
--- a/vllm/config.py
+++ b/vllm/config.py
@ -282,6 +282,10 @@ class ModelConfig:
            raise ValueError(
                "BitAndBytes quantization with TP or PP is not supported yet.")

+        if self.quantization == "bitsandbytes" and self.enforce_eager is False:
+            raise ValueError(
+                "BitAndBytes with enforce_eager = False is not supported yet.")
+
    def get_hf_config_sliding_window(self) -> Optional[int]:
        """Get the sliding window size, or None if disabled."""