Fix ci test "test_eval_fp8_accuracy" failed (#5185)

Co-authored-by: wunhuang <wunhuang@amd.com>

python/sglang/srt/layers/quantization/fp8_utils.py

@@ -243,8 +243,18 @@ def apply_fp8_linear(
         if _is_cuda:
             qinput, x_scale = sglang_per_token_quant_fp8(input_2d)
         else:
+            # TODO(kkhuang): temporarily enforce per-tensor activation scaling if weight is per-tensor scaling
+            # final solution should be: 1. add support to per-tensor activation scaling.
+            # 2. solve the torch.compile error from weight_scale.numel() == 1 and x_scale.numel() > 1 (below line#308)
+            if _is_hip and weight_scale.numel() == 1:
                 qinput, x_scale = ops.scaled_fp8_quant(
-                input_2d, input_scale, use_per_token_if_dynamic=use_per_token_if_dynamic
+                    input_2d,
+                    input_scale,
+                    use_per_token_if_dynamic=use_per_token_if_dynamic,
+                )
+            else:
+                qinput, x_scale = per_token_group_quant_fp8(
+                    input_2d, group_size=input_2d.shape[1]
                 )
 
     if cutlass_fp8_supported: