fix: quantize_activation_nvfp4 returns 2 values, not 3

2026-05-25 03:17:13 +00:00
parent 6dac3bcaf0
commit 1f310defa0
1 changed files with 1 additions and 1 deletions
--- a/tests/unit/test_nvfp4_quantize_kernel.py
+++ b/tests/unit/test_nvfp4_quantize_kernel.py
@@ -90,7 +90,7 @@ def test_nvfp4_quantize_correctness():
    x = torch.randn(M, N, dtype=torch.bfloat16, device='cuda')
    global_scale = 1.0
    
-    x_fp4, block_scale, _ = quantize_activation_nvfp4(x, global_scale)
+    x_fp4, block_scale = quantize_activation_nvfp4(x, global_scale)
    
    print(f"  Input: ({M}, {N}) BF16")
    print(f"  FP4 output: {x_fp4.shape}, dtype={x_fp4.dtype}")