[CI/Build] Avoid CUDA initialization (#8534)

2024-09-18 18:38:11 +08:00
parent e351572900
commit 6ffa3f314c
55 changed files with 256 additions and 256 deletions
--- a/vllm/model_executor/layers/quantization/fbgemm_fp8.py
+++ b/vllm/model_executor/layers/quantization/fbgemm_fp8.py
@@ -32,9 +32,7 @@ class FBGEMMFp8Config(QuantizationConfig):

        # For GPUs that lack FP8 hardware support, we can leverage the Marlin
        # kernel for fast weight-only FP8 quantization
-        capability = current_platform.get_device_capability()
-        capability = capability[0] * 10 + capability[1]
-        self.use_marlin = capability < 89
+        self.use_marlin = not current_platform.has_device_capability(89)

    @classmethod
    def get_name(cls) -> str: