static_scaled_fp8_quant should not run when scale.numel is not 1 (#20076)

8b8c209e · Eldar Kurtić · GitHub · 23a04e08 · 8b8c209e
Unverified Commit 8b8c209e authored Jun 25, 2025 by Eldar Kurtić Committed by GitHub Jun 25, 2025
Show whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

vllm/_custom_ops.py vllm/_custom_ops.py +1 -1

No files found.
--- a/vllm/_custom_ops.py
+++ b/vllm/_custom_ops.py
@@ -1276,7 +1276,7 @@ def scaled_fp8_quant(
            torch.ops._C.dynamic_scaled_fp8_quant(output, input, scale)
    else:
        # num_token_padding not implemented for this case
-        assert (scale.numel() == 1 or num_token_padding is None)
+        assert (scale.numel() == 1 and num_token_padding is None)
        torch.ops._C.static_scaled_fp8_quant(output, input, scale)

    return output, scale