default W4A16 alias to use group_size=128 (#94) (#95)

neuralmagic · Jun 20, 2024 · 42575fd · 42575fd
1 parent b77b1e6
commit 42575fd
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/src/compressed_tensors/quantization/quant_scheme.py b/src/compressed_tensors/quantization/quant_scheme.py
@@ -111,7 +111,7 @@ def is_preset_scheme(name: str) -> bool:
     weights=QuantizationArgs(), input_activations=QuantizationArgs(symmetric=True)
 )
 
-W4A16 = dict(weights=QuantizationArgs(num_bits=4, symmetric=True))
+W4A16 = dict(weights=QuantizationArgs(num_bits=4, group_size=128))
 
 PRESET_SCHEMES = {
     "W8A8": W8A8,