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

hot fix for nfp4 scheme (#784)

Wenhua Cheng7e52af2a17fc
diff --git a/auto_round/schemes.py b/auto_round/schemes.py
index e556acd29..496af179c 100644
--- a/auto_round/schemes.py
+++ b/auto_round/schemes.py
@@ -120,7 +120,7 @@ NVFP4 = QuantizationScheme.from_dict(
"group_size": 16,
"data_type": "nv_fp",
"act_bits": 4,
- "act_data_type": "nv_fp_with_static_gs",
+ "act_data_type": "nv_fp4_with_static_gs",
}
)