Skip to content

PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度

remove torch compile (#807)

Wenhua Cheng1452d95f4767
diff --git a/auto_round/data_type/nvfp.py b/auto_round/data_type/nvfp.py
index 12e8e4bfb..5d7e6221f 100644
--- a/auto_round/data_type/nvfp.py
+++ b/auto_round/data_type/nvfp.py
@@ -67,7 +67,6 @@ def calculate_gparam(tensor, group_size=16, device="cpu"):
return global_scale
def ref_nvfp4_quant(x, global_scale, block_size=16, v=0, scale_coeff=1.0):
assert global_scale.dtype == torch.float32
assert x.ndim == 2